WO2010101150A1 - Device for diagnosing pancreatic cystic disease - Google Patents
Device for diagnosing pancreatic cystic disease Download PDFInfo
- Publication number
- WO2010101150A1 WO2010101150A1 PCT/JP2010/053346 JP2010053346W WO2010101150A1 WO 2010101150 A1 WO2010101150 A1 WO 2010101150A1 JP 2010053346 W JP2010053346 W JP 2010053346W WO 2010101150 A1 WO2010101150 A1 WO 2010101150A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- cystic disease
- pancreatic cystic
- input
- pancreatic
- neural network
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
Definitions
- the present invention relates to a pancreatic cystic disease diagnostic apparatus used in medical institutions and the like, and more particularly to a pancreatic cystic disease diagnostic apparatus that diagnoses pancreatic cystic disease using an artificial neural network.
- pancreatic cystic diseases include intraductal papillary tumor IPMN, mucinous cystic tumor MCN, serous cystadenoma SCT, Solid-pseudopapillary tumor SPT, islet tumor, pancreatic acinar cell tumor, pancreatic pseudocyst, simple cyst, lymphoma and many others Diagnosis, including disease, is sometimes difficult, and doctors vary greatly in the rate of correct diagnosis.
- Diagnostic tools have been developed so far, but many of them have been diagnosed using the theory that the relationship between each event is assumed to be linear. However, since many of the events in the natural world are in a non-linear relationship, the correct diagnosis rate was not as high as expected. In addition, many of the tools were able to set only two items to be predicted, such as the presence or absence of disease and the strength of the disease state.
- the present invention provides a pancreatic cyst according to claim 1 simply by inputting patient data to each predetermined input item based on information obtained from clinical symptoms, findings, and various normal image examinations.
- An object of the present invention is to provide a pancreatic cystic disease diagnostic apparatus capable of automatically and accurately determining a sex disease name.
- An object of the present invention is to provide a pancreatic cystic disease diagnostic apparatus that can automatically and accurately determine the name of a pancreatic cystic disease simply by inputting data.
- An object of the present invention is to provide a pancreatic cystic disease diagnosis apparatus that can quickly and accurately determine the name of a pancreatic cystic disease simply by inputting patient data in the input items.
- the name of the pancreatic cystic disease can be obtained by inputting patient data in each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination. It is an object of the present invention to provide a pancreatic cystic disease diagnosis apparatus that can determine rapidly and accurately.
- the name of the pancreatic cystic disease can be obtained by inputting patient data into each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination.
- a device for diagnosing pancreatic cystic disease that can automatically and accurately determine an artificial neural network and an RBF network via an Internet line or the like to improve diagnostic accuracy is provided. The purpose is that.
- pancreatic cystic disease is used to determine the name of pancreatic cystic disease using one or more input data of clinical symptoms, findings, and image examinations.
- the artificial neural network is made to learn the contents of each input item and each disease name, and when the contents of each input item are input, the artificial neural network is operated to output the disease name It is characterized by.
- a neuron using a snake side function or a sigmoid function is used as a unit constituting the artificial neural network.
- a neuron using a radial basis function is used as a unit constituting the artificial neural network.
- the pancreatic cystic disease diagnosis is performed via any of a wired cable, a wireless line, and an Internet line. It is characterized in that the program is updated, the artificial neural network described in the pancreatic cystic disease diagnosis program, and the RBF network are learned.
- pancreatic cystic disease based on information obtained from clinical symptoms, findings and various normal image examinations, it is only necessary to input patient data into each predetermined input item.
- the name of the pancreatic cystic disease can be automatically and accurately determined.
- pancreatic cystic disease diagnosis apparatus by using an artificial neural network that is relatively easy to learn, while shortening the learning process period, clinical symptoms, findings and various normal image examinations can be performed. Based on the information obtained, the name of the pancreatic cystic disease can be automatically and accurately determined simply by inputting the patient data into each predetermined input item.
- pancreatic cystic disease diagnosis apparatus even when the input item content and the name of the pancreatic cystic disease are in a non-linear relationship, information obtained from clinical symptoms, findings, and various normal image examinations is included. Based on this, it is possible to automatically and accurately determine the name of the pancreatic cystic disease only by inputting the patient data to each predetermined input item.
- the patient image is simply input to each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination.
- the name of the pancreatic cystic disease can be automatically and accurately determined.
- the patient image is simply input to each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination.
- the patient image is simply input to each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination.
- FIG. 1 It is a block diagram which shows one form corresponding to Claims 1 and 2 among the pancreatic cystic disease diagnostic apparatuses by this invention. It is a schematic diagram which shows an example of the artificial neural network used with the pancreatic cystic disease diagnostic apparatus shown in FIG. It is a schematic diagram which shows an example of the neuron used with the artificial neural network shown in FIG. It is a graph which shows an example of the snake side function used with the neuron shown in FIG. It is a top view which shows an example of the input data table used with the pancreatic cystic disease diagnostic apparatus shown in FIG. It is a top view which shows an example of the correspondence table used with the pancreatic cystic disease diagnostic apparatus shown in FIG.
- FIG. 1 is a block diagram showing an embodiment corresponding to claims 1 and 2 of a pancreatic cystic disease diagnosis apparatus according to the present invention.
- the pancreatic cystic disease diagnostic apparatus 1a shown in this figure takes in a screen display by capturing display data, a keyboard 2 that is operated when inputting a patient's name, diagnosis content, and the like, a mouse 3 that is used as a pointing device. And a computer 5a for diagnosing pancreatic cystic disease using the installed pancreatic cystic disease diagnosis program, data output from the keyboard 2, point data output from the mouse 3, and the like.
- a computer 5a for diagnosing pancreatic cystic disease using the installed pancreatic cystic disease diagnosis program, data output from the keyboard 2, point data output from the mouse 3, and the like.
- the keyboard 2 is arranged in a keyboard case formed in a flat box shape, various character keys, numeric keys, various function keys arranged in an opening formed in the upper part of the keyboard case, and in the keyboard case. And an encoder that generates input data according to the operation content when a character key or function key is operated, and when the character key or function key is operated by a diagnostician or the like, the input data is It is generated and supplied to the computer 5.
- the mouse 3 is formed in a size that can be accommodated by a diagnostician's hand, and is movably disposed on the mouse pad.
- the mouse 3 is disposed in the mouse housing, and the mouse housing is moved.
- a movement detection mechanism that detects the movement of the mouse case by using a laser method using laser light, an optical method using infrared rays, and the like, and generates a movement direction data and a movement amount data. It is equipped with a button mechanism that generates a right click signal, a left click signal, etc. when it is placed and pressed by a diagnostician's finger, etc., and movement direction data when moved on the mouse pad by a diagnostician
- the movement amount data is generated and supplied to the computer 5. Further, when the button mechanism is operated by a finger of a diagnostician or the like, a right click signal or a left click signal is generated and supplied to the computer 5.
- the display 4 is provided on a display casing formed in a flat box shape, a back surface of the display casing, a support base that supports the display casing so as not to fall down, and a front surface of the display casing.
- Display data output from the computer 5 is provided with a liquid crystal panel disposed in the opening, and a decoder disposed in the display housing, which decodes display data output from the computer and displays the data on the liquid crystal panel. And display on the LCD panel.
- the computer 5a captures input data output from the keyboard 2 while supplying a power supply voltage to the keyboard 2, converts it into parallel data of a specified format, and sends the parallel data to the system bus 6. While supplying the power supply voltage to the mouse 3, the serial movement direction data and movement amount data output from the mouse 3 are captured and converted into parallel movement direction data and movement amount data on the system bus 6.
- the hard disk mechanism 14a for taking in and storing the data via the system bus 6.
- the keyboard 2 and mouse 3 are operated in a state in which the pancreatic cystic disease diagnosis program is started, and the patient is assigned to each input item displayed on the display 4.
- a diagnosis start button displayed on the display 4 is clicked, a pancreatic cystic disease is diagnosed and displayed on the display 4 by the artificial neural network 15 using a multilayer perceptron.
- the artificial neural network 15 includes a plurality of neurons 16, each of which receives a plurality of input signals corresponding to the contents of input items specified in advance, and outputs a plurality of output signals. 17 and a plurality of neurons 16, which captures the output signal of each neuron 16 constituting the input layer 17 and outputs a plurality of output signals, and a plurality of neurons 16.
- An output layer 19 that captures an output signal of each neuron 16 constituting the device and outputs a plurality of output signals, and before starting diagnosis of pancreatic cystic disease, a lot of clinical data is used to perform learning processing. And the value of the weight coefficient “Wi” of each neuron 16 is optimized.
- Each neuron 16 has a plurality of input terminals and one output terminal as shown in FIG. 3, and performs the operations shown in the following equations (1) and (2) to obtain each input signal “Xi”,
- a function that calculates the multiplication value “Wi * Xi” with each weight coefficient “Wi” and subtracts a preset threshold value “ ⁇ ” from the sum “ ⁇ Wi * Xi” of each multiplication value “Wi * Xi” “F ( ⁇ Wi * Xi- ⁇ )” is treated as the snake side function shown in FIG. 4, and if “f ( ⁇ Wi * Xi- ⁇ ) ⁇ 0”, the value “1” is output as the output signal “y”.
- pancreatic cystic disease diagnosis apparatus 1a Next, the operation of the pancreatic cystic disease diagnosis apparatus 1a will be described with reference to the block diagram shown in FIG. 1, the schematic diagrams shown in FIGS. 2 and 3, and the graph shown in FIG.
- ⁇ Learning action> First, prior to the actual diagnosis, patient data on specific pancreatic cystic disease was extracted from each pancreatic cystic disease patient by a specialist based on information obtained from clinical findings and various imaging tests. Thereafter, patient data is organized for each patient with pancreatic cystic disease, data corresponding to the input items shown below is organized for each patient, and an input data table 20 shown in FIG. 5 is created. (1) Age, gender, (2) Cys shape, number, wall thickness, solid component, calcification and its parts, (3) Imaging conditions, (4) Signal intensity between cysts, traffic with the main pancreatic duct, (5) Expansion of the main pancreatic duct At this time, as data corresponding to each input item, one of “0”, “1”, and “2” can be selected. The content is easy to input.
- pancreatic cystic disease person Further, patient data is organized for each pancreatic cystic disease person, and it is diagnosed which pancreatic cystic disease name each patient corresponds to, for example, the name of the pancreatic cystic disease as shown below.
- a correspondence table 21 as shown is created.
- IPMN Intraductal papillary tumor
- MCN Myxoid cyst tumor
- SCT Serous cystadenoma SCT
- Solid-pseudopapillary tumor SPT (5) islet tumor, (6) Pancreatic acinar cell tumor, (7) Pancreatic pseudocyst, (8) simple cyst, (9) Lymphoma
- each input item data is assigned to each input signal “Xi” of the input layer 17, and each pancreatic cystic disease name is assigned to each output signal “ym” of the output layer 19. .
- each weight coefficient “Wji (t + 1)” is optimized by the gradient method, the least square method, or the like as follows.
- Wji (t + 1) Wji (t) + C (tj-yj) * Xi (3)
- C learning coefficient j: number specifying the input layer 17, intermediate layer 18, and output layer 19
- i number specifying the input signal t: learning number tj: teacher signal
- yj output signal
- each weight coefficient “Wji (t + 1)” is optimized by the gradient method, the least square method, and the like as shown in the equation (3).
- each weighting factor “Wji (t + 1)” of the intermediate layer 18 is not updated, and only each weighting factor “Wji (t + 1)” of the output layer 19 is updated. Is called.
- an input data table 20 corresponding to input items as shown in FIG. 6 is created by a diagnostician based on information obtained from clinical findings and various image examinations.
- the check box corresponding to the input data table 20 is selected from the three check boxes for each input item displayed on the display 4 by operating the keyboard 2 and the mouse 3 by the diagnostician. Is selected and the “re” point is entered.
- the computer 5a causes the learned artificial neural network (with the installed pancreatic cystic disease diagnosis program).
- the described artificial neural network) 15 is activated, and the name of the pancreatic cystic disease corresponding to the content of the input item is selected and displayed on the display 4.
- a predetermined number of input items are used, and patient data is input in an alternative selection format from “0”, “1”, and “2” for each input item.
- diagnosis start button is clicked, the artificial neural network 15 trained in advance is operated to display the name of the pancreatic cystic disease corresponding to the patient data, so that the learning processing period is shortened.
- the patient's data is entered into each predetermined input item, and the name of the pancreatic cystic disease is automatically and accurately determined. (Effects of claims 1 and 2).
- the snake side function is used, but other functions such as a sigmoid function may be used.
- FIG. 7 is a block diagram showing an embodiment corresponding to claim 3 of the pancreatic cystic disease diagnosis apparatus according to the present invention.
- parts corresponding to those in FIG. 1 are denoted by the same reference numerals.
- the pancreatic cystic disease diagnostic apparatus 1b shown in this figure is different from the pancreatic cystic disease diagnostic apparatus 1a shown in FIG. 1 in that a radial basis function (RBF) type artificial neural network (RBF network) 31 ( 8), a pancreatic cystic disease diagnosis program is created and stored in the program area 13b of the hard disk mechanism 14b.
- a radial basis function (RBF) type artificial neural network (RBF network) 31 ( 8) a pancreatic cystic disease diagnosis program is created and stored in the program area 13b of the hard disk mechanism 14b.
- the RBF network 31 includes a plurality of units 32, each of which receives a plurality of input signals corresponding to the contents of input items specified in advance, and outputs a plurality of output signals.
- the unit 32 includes the intermediate layer 34 that takes in the output signal of each unit 32 that constitutes the input layer 33 and outputs a plurality of output signals, and each unit that comprises the plurality of units 32 and constitutes the intermediate layer 34.
- An output layer 35 that captures 32 output signals and outputs a plurality of output signals.
- the keyboard 2 and the mouse 3 are operated, the patient test results are input to each input item displayed on the display 4, and the input layer 33 is configured when the diagnosis start button is clicked.
- An output signal corresponding to the test result is output from each unit 32, processed by each unit 32 configuring the intermediate layer 34, and an output signal corresponding to the diagnosis result is output from each unit 32 configuring the output layer 35. .
- Each unit 32 has a plurality of input terminals and one output terminal as shown in FIG. 9, and is represented by a radial basis function corresponding to each input signal “x”, for example, the following equation (4):
- a Gaussian function operation is performed in which a maximum value (or minimum value) is taken at the center and monotonously decreasing (or increasing) as the distance from the center increases, and each radial basis function “ ⁇ i ( x) ”and using the following equation (5), the product of each radial basis function“ ⁇ i (x) ”and each weight coefficient“ Wi ”is multiplied by“ Wi * ⁇ i (x ) ”And add the preset bias value“ w0 ”to the sum“ ⁇ Wi * ⁇ i (x) ”of each multiplication value“ Wi * ⁇ i (x) ”to obtain the output function“ y ”.
- ⁇ i (x) (x-ci) T (x-ci) / (2 ⁇ 2i) (4)
- ⁇ i (x): i-th radial basis function i: value indicating input signal, radial basis function number (i 1, 2,..., N)
- ci Center of i-th radial basis function (Gaussian function)
- T Symbol for transposing the vector ⁇ : Standard deviation of Gaussian function, parameter that determines width x: Input signal x-ci: Input signal and Gaussian function center Euclidean distance (vector notation)
- the RBF network 31 configured as described above to perform the learning operation described above, the data for each input item illustrated in FIG. 5 and the diagnosis result illustrated in FIG. Even if it is non-linear, the patient's pancreatic cystic disease name is automatically entered by simply entering patient data into each predetermined entry based on information obtained from clinical symptoms, findings and various normal imaging tests. And can be determined accurately (effect of claim 3).
- FIG. 11 is a block diagram showing an embodiment corresponding to claim 4 of the pancreatic cystic disease diagnosis apparatus according to the present invention.
- parts corresponding to those in FIG. 1 are denoted by the same reference numerals.
- the pancreatic cystic disease diagnostic apparatus 1c shown in this figure is different from the pancreatic cystic disease diagnostic apparatus 1a shown in FIG. 1 in that a communication circuit 41 is provided in the computer 5c, and communication is performed via a hub (not shown). That is, the circuit 41 and a LAN cable (Ethernet cable or wireless LAN line) 42 are connected.
- a communication circuit 41 is provided in the computer 5c, and communication is performed via a hub (not shown). That is, the circuit 41 and a LAN cable (Ethernet cable or wireless LAN line) 42 are connected.
- CT Computer tomography
- MRI magnetic resonance imaging
- EUS endoscopic ultrasonography
- ERCP endoscopic retrograde cholangio-pancreatography
- the diagnosis result is directly captured from the apparatus 47 and displayed on the display 4, and when the keyboard 2 and mouse 3 are operated to input the analysis result analysis instruction, the diagnosis result is stored in the program area 13c. Execute the diagnostic result analysis program to quantize the diagnostic result of the US device 43 to the diagnostic result of the ERCP device 47 into “0”, “1”, “2”, and then process it as data of each input item It is to do so.
- FIG. 12 is a block diagram showing an embodiment corresponding to claim 5 of the pancreatic cystic disease diagnosis apparatus according to the present invention. In this figure, parts corresponding to those in FIG. 11 are given the same reference numerals.
- the pancreatic cystic disease diagnostic apparatus 1d shown in this figure is different from the pancreatic cystic disease diagnostic apparatus 1c shown in FIG. 1 in that a gateway device (or router device) 51 for connecting the LAN cable 42 and the Internet line is provided.
- the remote program is stored in the program area 13d of the hard disk mechanism 14d, and the mote program is operated by a server device (not shown) connected to the Internet to update the pancreatic cystic disease diagnostic program,
- the artificial neural network 15 described in the sexually transmitted disease diagnosis program is learned.
- the server device updates the pancreatic cystic disease diagnosis program or learns the artificial neural network 15 (or RBF network 31) described in the pancreatic cystic disease diagnosis program, Cystic disease diagnosis accuracy can be further improved (effect of claim 5).
- the diagnosis is performed using the artificial neural network 15 and the RBF network 31, so that the patient's simple input items are included.
- the name of the pancreatic cystic disease can be determined with a high diagnosis rate.
- the prediction efficiency can be further improved by accumulating the number of cases and setting more detailed input items.
- Such a diagnostic method using the artificial neural network 15 and the RBF network 31 can be applied not only to image diagnosis of pancreatic cystic disease but also to general disease diagnosis, and can be a new diagnostic modality.
- the present invention relates to a pancreatic cystic disease diagnostic apparatus used in medical institutions, and more particularly to a pancreatic cystic disease diagnostic apparatus that diagnoses pancreatic cystic disease using an artificial neural network, and has industrial applicability.
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
A device whereby a pancreatic cystic disease name can be automatically and accurately determined simply by inputting, based on the information obtained from clinical symptoms, findings and various imaging tests commonly employed, patient's data in accordance with preset input items. By using a preset number of input items, patient's data is input for each input item with a threefold choice, i.e., "0"/"1"/"2". When a diagnosis-start button is clicked, an artificial neural network (15), which has preliminarily learned, is activated and a pancreatic cystic disease name corresponding to the patient's data is displayed.
Description
本発明は、医療機関などで使用される膵嚢胞性疾患診断装置に係わり、特に人工ニューラルネットワークを使用して、膵嚢胞性疾患を診断する膵嚢胞性疾患診断装置に関する。
The present invention relates to a pancreatic cystic disease diagnostic apparatus used in medical institutions and the like, and more particularly to a pancreatic cystic disease diagnostic apparatus that diagnoses pancreatic cystic disease using an artificial neural network.
健康診断での腹部エコー検査の普及と、検出精度の向上により、膵嚢胞性疾患患者が頻繁に消化器内科の外来を受診するようになった。膵嚢胞性疾患は、膵管内乳頭腫瘍ＩＰＭＮ、粘液性嚢胞腫瘍ＭＣＮ、漿液性嚢胞腺腫ＳＣＴ、Solid-pseudopapillary tumor ＳＰＴ、膵島腫瘍、膵腺房細胞腫瘍、膵仮性嚢胞、単純嚢胞、リンパ腫など多くの疾患を含み、その診断は時に困難であり、医師により正診率に大きな差がある。
普及 With the widespread use of abdominal echocardiography in health checkups and improved detection accuracy, patients with pancreatic cystic diseases have frequently visited outpatients in the gastroenterology department. Pancreatic cystic diseases include intraductal papillary tumor IPMN, mucinous cystic tumor MCN, serous cystadenoma SCT, Solid-pseudopapillary tumor SPT, islet tumor, pancreatic acinar cell tumor, pancreatic pseudocyst, simple cyst, lymphoma and many others Diagnosis, including disease, is sometimes difficult, and doctors vary greatly in the rate of correct diagnosis.
また、日常臨床においては、忙殺された時間の中で十分な診断的検討がなされないまま臨床的判断と対応を迫られる場面に頻回に遭遇する。
Also, in daily clinical practice, we often encounter scenes where we are forced to make clinical judgments and responses without being fully diagnosed during a busy time.
このような状況を打開するために、臨床症状、所見と種々の通常の画像検査から得られる情報をもとに、膵嚢胞性疾患名などを判断する診断ツールが開発されている。
In order to overcome this situation, diagnostic tools have been developed to determine the names of pancreatic cystic diseases based on clinical symptoms, findings, and information obtained from various normal imaging tests.
これまでも診断ツールは開発されてきたが、その多くは各事象間の関係が線形であると仮定した理論を用いて診断していた。しかしながら、自然界の事象の多くは非線形な関係にあるため、その正診率は期待されるほど高くはなかった。また、ツールの多くが疾患の有無や病態の程度の強弱など、予測する項目を２段階までしか設定できなかった。
Diagnostic tools have been developed so far, but many of them have been diagnosed using the theory that the relationship between each event is assumed to be linear. However, since many of the events in the natural world are in a non-linear relationship, the correct diagnosis rate was not as high as expected. In addition, many of the tools were able to set only two items to be predicted, such as the presence or absence of disease and the strength of the disease state.
本発明は上記の事情に鑑み、請求項１では、臨床症状、所見と種々の通常の画像検査から得られる情報に基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができる膵嚢胞性疾患診断装置を提供することを目的としている。
In view of the above circumstances, the present invention provides a pancreatic cyst according to claim 1 simply by inputting patient data to each predetermined input item based on information obtained from clinical symptoms, findings, and various normal image examinations. An object of the present invention is to provide a pancreatic cystic disease diagnostic apparatus capable of automatically and accurately determining a sex disease name.
また、請求項２では、学習処理が比較的容易な人工ニューラルネットワークを使用することにより、臨床症状、所見と種々の通常の画像検査から得られる情報に基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができる膵嚢胞性疾患診断装置を提供することを目的としている。
Further, in claim 2, by using an artificial neural network that is relatively easy to learn, a patient is assigned to each predetermined input item based on information obtained from clinical symptoms, findings, and various normal image examinations. An object of the present invention is to provide a pancreatic cystic disease diagnostic apparatus that can automatically and accurately determine the name of a pancreatic cystic disease simply by inputting data.
また、請求項３では、入力項目内容と、膵嚢胞性疾患名とが非線形な関係にある場合でも、臨床症状、所見と種々の通常の画像検査から得られる情報に基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を迅速に、かつ正確に判定することができる膵嚢胞性疾患診断装置を提供することを目的としている。
Further, in claim 3, even when the input item content and the name of the pancreatic cystic disease are in a non-linear relationship, each of the predetermined items is determined based on clinical symptoms, findings, and information obtained from various normal image examinations. An object of the present invention is to provide a pancreatic cystic disease diagnosis apparatus that can quickly and accurately determine the name of a pancreatic cystic disease simply by inputting patient data in the input items.
また、請求項４では、通常の画像検査を自動入力させながら、臨床症状、所見と種々の情報などに基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を、迅速にかつ正確に判定することができる膵嚢胞性疾患診断装置を提供することを目的としている。
Further, according to claim 4, the name of the pancreatic cystic disease can be obtained by inputting patient data in each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination. It is an object of the present invention to provide a pancreatic cystic disease diagnosis apparatus that can determine rapidly and accurately.
また、請求項５では、通常の画像検査を自動入力させながら、臨床症状、所見と種々の情報などに基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができるとともに、インターネット回線などを介して、人工ニューラルネットワーク、ＲＢＦネットワークを学習させて、診断精度を向上させることができる膵嚢胞性疾患診断装置を提供することを目的としている。
Further, according to claim 5, the name of the pancreatic cystic disease can be obtained by inputting patient data into each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination. A device for diagnosing pancreatic cystic disease that can automatically and accurately determine an artificial neural network and an RBF network via an Internet line or the like to improve diagnostic accuracy is provided. The purpose is that.
上記の目的を達成するために本発明は、請求項１では、臨床症状、所見、各画像検査のいずれか１つ以上の入力データを用いて、膵嚢胞性疾患名を判定する膵嚢胞性疾患診断装置において、人工ニューラルネットワークに、各入力項目の内容と、各疾患名とを学習させ、各入力項目の内容が入力されたとき、前記人工ニューラルネットワークを動作させて、疾患名を出力させることを特徴としている。
In order to achieve the above object, according to the present invention, in claim 1, pancreatic cystic disease is used to determine the name of pancreatic cystic disease using one or more input data of clinical symptoms, findings, and image examinations. In the diagnostic device, the artificial neural network is made to learn the contents of each input item and each disease name, and when the contents of each input item are input, the artificial neural network is operated to output the disease name It is characterized by.
また、請求項２では、請求項１に記載の膵嚢胞性疾患診断装置において、前記人工ニューラルネットワークを構成するユニットとして、ヘビサイド関数、またはシグモイド関数を使用したニューロンを使用することを特徴としている。
Further, in claim 2, in the pancreatic cystic disease diagnosis apparatus according to claim 1, a neuron using a snake side function or a sigmoid function is used as a unit constituting the artificial neural network.
また、請求項３では、請求項１に記載の膵嚢胞性疾患診断装置において、前記人工ニューラルネットワークを構成するユニットとして、動径基底関数を使用したニューロンを使用することを特徴としている。
Further, in claim 3, in the pancreatic cystic disease diagnosis apparatus according to claim 1, a neuron using a radial basis function is used as a unit constituting the artificial neural network.
また、請求項４では、請求項１、２、３のいずれかに記載の膵嚢胞性疾患診断装置において、有線ケーブル、または無線回線を介して、ＵＳ装置、ＣＴ装置、ＭＲＩ装置、ＥＵＳ装置、ＥＲＣＰ装置から、直接、診断結果を取り込むことを特徴としている。
Moreover, in Claim 4, in the pancreatic cystic disease diagnostic apparatus according to any one of Claims 1, 2, and 3, via a wired cable or a wireless line, a US apparatus, a CT apparatus, an MRI apparatus, an EUS apparatus, It is characterized in that a diagnostic result is taken directly from the ERCP device.
また、請求項５では、請求項１、２、３、４のいずれかに記載の膵嚢胞性疾患診断装置において、有線ケーブル、無線回線、インターネット回線のいずれかを介して、膵嚢胞性疾患診断プログラムが更新される処理、膵嚢胞性疾患診断プログラムで記述されている人工ニューラルネットワーク、ＲＢＦネットワークの学習処理が行われることを特徴としている。
Further, in claim 5, in the pancreatic cystic disease diagnosis apparatus according to any one of claims 1, 2, 3, and 4, the pancreatic cystic disease diagnosis is performed via any of a wired cable, a wireless line, and an Internet line. It is characterized in that the program is updated, the artificial neural network described in the pancreatic cystic disease diagnosis program, and the RBF network are learned.
本発明による、請求項１の膵嚢胞性疾患診断装置では、臨床症状、所見と種々の通常の画像検査から得られる情報に基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができる。
In the diagnostic apparatus for pancreatic cystic disease according to the present invention according to the present invention, based on information obtained from clinical symptoms, findings and various normal image examinations, it is only necessary to input patient data into each predetermined input item. The name of the pancreatic cystic disease can be automatically and accurately determined.
また、請求項２の膵嚢胞性疾患診断装置では、学習処理が比較的容易な人工ニューラルネットワークを使用することにより、学習処理期間を短縮させながら、臨床症状、所見と種々の通常の画像検査から得られる情報に基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができる。
Further, in the pancreatic cystic disease diagnosis apparatus according to claim 2, by using an artificial neural network that is relatively easy to learn, while shortening the learning process period, clinical symptoms, findings and various normal image examinations can be performed. Based on the information obtained, the name of the pancreatic cystic disease can be automatically and accurately determined simply by inputting the patient data into each predetermined input item.
また、請求項３の膵嚢胞性疾患診断装置では、入力項目内容と、膵嚢胞性疾患名とが非線形な関係にある場合でも、臨床症状、所見と種々の通常の画像検査から得られる情報に基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができる。
Further, in the pancreatic cystic disease diagnosis apparatus according to claim 3, even when the input item content and the name of the pancreatic cystic disease are in a non-linear relationship, information obtained from clinical symptoms, findings, and various normal image examinations is included. Based on this, it is possible to automatically and accurately determine the name of the pancreatic cystic disease only by inputting the patient data to each predetermined input item.
また、請求項４の膵嚢胞性疾患診断装置では、通常の画像検査を自動入力させながら、臨床症状、所見と種々の情報などに基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができる。
In the pancreatic cystic disease diagnosis apparatus according to claim 4, the patient image is simply input to each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination. Thus, the name of the pancreatic cystic disease can be automatically and accurately determined.
また、請求項５の膵嚢胞性疾患診断装置では、通常の画像検査を自動入力させながら、臨床症状、所見と種々の情報などに基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができるとともに、インターネット回線などを介して、人工ニューラルネットワーク、ＲＢＦネットワークを学習させて、診断精度を向上させることができる。
In the pancreatic cystic disease diagnosis apparatus according to claim 5, the patient image is simply input to each predetermined input item based on clinical symptoms, findings and various information while automatically inputting a normal image examination. Thus, it is possible to automatically and accurately determine the name of the pancreatic cystic disease, and to learn the artificial neural network and the RBF network via the Internet line or the like, thereby improving the diagnostic accuracy.
《第１形態》
図１は本発明による膵嚢胞性疾患診断装置のうち、請求項１、２に対応する一形態を示すブロック図である。 << First form >>
FIG. 1 is a block diagram showing an embodiment corresponding to claims 1 and 2 of a pancreatic cystic disease diagnosis apparatus according to the present invention.
図１は本発明による膵嚢胞性疾患診断装置のうち、請求項１、２に対応する一形態を示すブロック図である。 << First form >>
FIG. 1 is a block diagram showing an embodiment corresponding to
この図に示す膵嚢胞性疾患診断装置１ａは、患者の氏名、診断内容などを入力するときに操作されるキーボード２と、ポインティングデバイスとして使用されるマウス３と、表示データを取り込んで、画面表示するディスプレイ４と、インストールされている膵嚢胞性疾患診断プログラム、キーボード２から出力されるデータ、マウス３から出力されるポイントデータなどを用いて、膵嚢胞性疾患診断を行うコンピュータ５ａとを備えており、キーボード２、マウス３が操作されて、ディスプレイ４上に表示されている各入力項目に患者の検査結果が入力され、ディスプレイ４上に表示されている診断開始ボタンがクリックされたとき、多層パーセプトロンを用いた人工ニューラルネットワーク１５（図２参照）を使用して、膵嚢胞性疾患を診断し、診断結果をディスプレイ４上に表示する。
The pancreatic cystic disease diagnostic apparatus 1a shown in this figure takes in a screen display by capturing display data, a keyboard 2 that is operated when inputting a patient's name, diagnosis content, and the like, a mouse 3 that is used as a pointing device. And a computer 5a for diagnosing pancreatic cystic disease using the installed pancreatic cystic disease diagnosis program, data output from the keyboard 2, point data output from the mouse 3, and the like. When the keyboard 2 and the mouse 3 are operated, patient test results are input to the input items displayed on the display 4, and the diagnosis start button displayed on the display 4 is clicked, multiple layers are displayed. Pancreatic cystic disease using artificial neural network 15 (see FIG. 2) with perceptron Diagnose, displays the diagnostic results on the display 4.
キーボード２は、平箱状に形成されるキーボード筐体と、キーボード筐体の上部に形成された開口部に配置される各種の文字キー、テンキー、各種のファンクションキーと、キーボード筐体内に配置され、文字キー又はファンクションキーなどが操作されたとき、操作内容に応じた入力データを生成するエンコーダとを備えており、診断医などによって、文字キー又はファンクションキーなどが操作されたとき、入力データを生成して、コンピュータ５に供給する。
The keyboard 2 is arranged in a keyboard case formed in a flat box shape, various character keys, numeric keys, various function keys arranged in an opening formed in the upper part of the keyboard case, and in the keyboard case. And an encoder that generates input data according to the operation content when a character key or function key is operated, and when the character key or function key is operated by a diagnostician or the like, the input data is It is generated and supplied to the computer 5.
また、マウス３は、診断医の手に収まる程度の大きさに形成され、マウスパッド上に移動自在に配置されるマウス筐体と、マウス筐体内に配置され、マウス筐体が動かされたとき、レーザー光を使用したレーザー方式、赤外線などを使用した光学方式などで、マウス筐体の動きを検知して、移動方向データ、移動量データを生成する移動量検知機構と、マウス筐体上に配置され、診断医の指などで押されたとき、右クリック信号、左クリック信号などを生成するボタン機構とを備えており、診断医などによって、マウスパッド上を動かされたとき、移動方向データ、移動量データを生成して、コンピュータ５に供給する。また、診断医の指などによって、ボタン機構が操作されたとき、右クリック信号、または左クリック信号などを生成して、コンピュータ５に供給する。
In addition, the mouse 3 is formed in a size that can be accommodated by a diagnostician's hand, and is movably disposed on the mouse pad. The mouse 3 is disposed in the mouse housing, and the mouse housing is moved. A movement detection mechanism that detects the movement of the mouse case by using a laser method using laser light, an optical method using infrared rays, and the like, and generates a movement direction data and a movement amount data. It is equipped with a button mechanism that generates a right click signal, a left click signal, etc. when it is placed and pressed by a diagnostician's finger, etc., and movement direction data when moved on the mouse pad by a diagnostician The movement amount data is generated and supplied to the computer 5. Further, when the button mechanism is operated by a finger of a diagnostician or the like, a right click signal or a left click signal is generated and supplied to the computer 5.
また、ディスプレイ４は、平箱状に形成されるディスプレイ筐体と、ディスプレイ筐体の背面に設けられ、ディスプレイ筐体が倒れないように、支持する支持台と、ディスプレイ筐体の前面に形成された開口部に配置される液晶パネルと、ディスプレイ筐体内に配置され、コンピュータから出力された表示データをデコードして、液晶パネルに表示させるデコーダとを備えており、コンピュータ５から出力される表示データを取り込み、液晶パネルに表示させる。
In addition, the display 4 is provided on a display casing formed in a flat box shape, a back surface of the display casing, a support base that supports the display casing so as not to fall down, and a front surface of the display casing. Display data output from the computer 5 is provided with a liquid crystal panel disposed in the opening, and a decoder disposed in the display housing, which decodes display data output from the computer and displays the data on the liquid crystal panel. And display on the LCD panel.
コンピュータ５ａは、キーボード２に電源電圧を供給しながら、キーボード２から出力される入力データを取り込み、指定された形式のパラレルデータに変換して、システムバス６上に送出するキーボードインタフェース回路７と、マウス３に電源電圧を供給しながら、マウス３から出力されるシリアル形式の移動方向データ、移動量データなどを取り込み、パラレル形式の移動方向データ、移動量データなどに変換し、システムバス６上に送出するマウスインタフェース回路８と、システムバス６から供給される表示データを取り込み、指定された規格の表示データに変換し、ディスプレイ４に供給するディスプレイインタフェース回路９と、各種のデータ処理を行うＣＰＵ回路１０と、ＤＤＲ－ＲＡＭ素子などによって構成され、ＣＰＵ回路１０の作業エリアなどとして使用されるメモリ回路１１と、ＣＰＵ回路１０を動作させるのに必要なＯＳ（Operating System）が格納されるＯＳエリア１２、膵嚢胞性疾患診断プログラムが格納されるプログラムエリア１３ａなどを持ち、ＣＰＵ回路１０から読み出し信号が出力されたとき、指定されたデータ、プログラムなどを読み出し、システムバス６を介して、ＣＰＵ回路１０に供給し、またＣＰＵ回路１０から書き込み信号が出力されたとき、システムバス６を介して、データを取り込み、記憶するハードディスク機構１４ａとを備えている。
The computer 5a captures input data output from the keyboard 2 while supplying a power supply voltage to the keyboard 2, converts it into parallel data of a specified format, and sends the parallel data to the system bus 6. While supplying the power supply voltage to the mouse 3, the serial movement direction data and movement amount data output from the mouse 3 are captured and converted into parallel movement direction data and movement amount data on the system bus 6. A mouse interface circuit 8 for sending out, a display interface circuit 9 for taking in display data supplied from the system bus 6 and converting it into display data of a specified standard, and supplying it to the display 4; and a CPU circuit for performing various data processing 10 and DDR-RAM elements, etc., and CP A memory circuit 11 used as a work area of the circuit 10, an OS area 12 in which an OS (Operating System) necessary for operating the CPU circuit 10 is stored, and a program area in which a pancreatic cystic disease diagnosis program is stored 13a and the like, and when a read signal is output from the CPU circuit 10, the designated data, program, etc. are read out and supplied to the CPU circuit 10 via the system bus 6, and a write signal is output from the CPU circuit 10. The hard disk mechanism 14a for taking in and storing the data via the system bus 6.
そして、コンピュータ５ａの電源がオンされた後、膵嚢胞性疾患診断プログラムが立ち上げられている状態で、キーボード２、マウス３が操作されて、ディスプレイ４上に表示されている各入力項目に患者の検査結果が入力され、ディスプレイ４上に表示されている診断開始ボタンがクリックされたとき、多層パーセプトロンを用いた人工ニューラルネットワーク１５で、膵嚢胞性疾患を診断し、ディスプレイ４上に表示する。
Then, after the computer 5a is turned on, the keyboard 2 and mouse 3 are operated in a state in which the pancreatic cystic disease diagnosis program is started, and the patient is assigned to each input item displayed on the display 4. When a diagnosis start button displayed on the display 4 is clicked, a pancreatic cystic disease is diagnosed and displayed on the display 4 by the artificial neural network 15 using a multilayer perceptron.
この場合、図２に示す如く人工ニューラルネットワーク１５は、複数のニューロン１６によって構成され、予め指定された入力項目の内容に対応する複数の入力信号を各々取り込み、複数の出力信号を出力する入力層１７と、複数のニューロン１６によって構成され、入力層１７を構成する各ニューロン１６の出力信号を取り込み、複数の出力信号を出力する中間層１８と、複数のニューロン１６によって構成され、中間層１８を構成する各ニューロン１６の出力信号を取り込み、複数の出力信号を出力する出力層１９とを備えており、膵嚢胞性疾患を診断を開始する前に、多数の臨床データが使用されて、学習処理が行われ、各ニューロン１６の重み係数“Wi”の値が最適化される。そして、診断時には、キーボード２、マウス３が操作されて、ディスプレイ４上に表示されている各入力項目に患者の検査結果が入力され、診断開始ボタンがクリックされたとき、入力層１７、中間層１８を構成する各ニューロン１６から検査結果に対応した出力信号が出力されて、出力層１９を構成する各ニューロン１６から診断結果に対応した出力信号が出力される。
In this case, as shown in FIG. 2, the artificial neural network 15 includes a plurality of neurons 16, each of which receives a plurality of input signals corresponding to the contents of input items specified in advance, and outputs a plurality of output signals. 17 and a plurality of neurons 16, which captures the output signal of each neuron 16 constituting the input layer 17 and outputs a plurality of output signals, and a plurality of neurons 16. An output layer 19 that captures an output signal of each neuron 16 constituting the device and outputs a plurality of output signals, and before starting diagnosis of pancreatic cystic disease, a lot of clinical data is used to perform learning processing. And the value of the weight coefficient “Wi” of each neuron 16 is optimized. At the time of diagnosis, when the keyboard 2 and mouse 3 are operated, the patient test results are input to the input items displayed on the display 4, and the diagnosis start button is clicked, the input layer 17, the intermediate layer An output signal corresponding to the test result is output from each neuron 16 configuring the output 18, and an output signal corresponding to the diagnosis result is output from each neuron 16 configuring the output layer 19.
各ニューロン１６は、図３に示すように複数の入力端子と１つの出力端子とを持ち、下記の（１）式、（２）式に示す演算を行って、各入力信号“Xi”と、各重み係数“Wi”との乗算値“Wi*Xi”を求めるとともに、各乗算値“Wi*Xi”の総和“ΣWi*Xi”から予め設定されているしきい値“θ”を減算した関数“f(ΣWi*Xi-θ)”を図４に示すヘビサイド関数として扱い、“f(ΣWi*Xi-θ)≧0”であれば、出力信号“y”として、値“１”を出力し、また“f(ΣWi*Xi-θ)＜0”であれば、出力信号“y”として、値“０”を出力する。
但し、f(x)：ヘビサイド関数
i：入力信号の番号を示す値（i=1,2,...,N）
Xi：i番目の入力信号の値
Wi：i番目の入力信号に対する重み係数
θ：しきい値
但し、y：出力信号の値
f(x)：ヘビサイド関数値“x”に対する出力信号の値 Eachneuron 16 has a plurality of input terminals and one output terminal as shown in FIG. 3, and performs the operations shown in the following equations (1) and (2) to obtain each input signal “Xi”, A function that calculates the multiplication value “Wi * Xi” with each weight coefficient “Wi” and subtracts a preset threshold value “θ” from the sum “ΣWi * Xi” of each multiplication value “Wi * Xi” “F (ΣWi * Xi-θ)” is treated as the snake side function shown in FIG. 4, and if “f (ΣWi * Xi-θ) ≧ 0”, the value “1” is output as the output signal “y”. If “f (ΣWi * Xi−θ) <0”, the value “0” is output as the output signal “y”.
Where f (x): snake side function i: value indicating the number of the input signal (i = 1,2, ..., N)
Xi: i-th input signal value Wi: weighting factor for i-th input signal θ: threshold
Where y: Output signal value f (x): Output signal value for snake side function value “x”
i：入力信号の番号を示す値（i=1,2,...,N）
Xi：i番目の入力信号の値
Wi：i番目の入力信号に対する重み係数
θ：しきい値
f(x)：ヘビサイド関数値“x”に対する出力信号の値 Each
Xi: i-th input signal value Wi: weighting factor for i-th input signal θ: threshold
次に、図１に示すブロック図、図２、図３に示す模式図、図４に示すグラフを参照しながら、膵嚢胞性疾患診断装置１ａの動作を説明する。
Next, the operation of the pancreatic cystic disease diagnosis apparatus 1a will be described with reference to the block diagram shown in FIG. 1, the schematic diagrams shown in FIGS. 2 and 3, and the graph shown in FIG.
＜学習動作＞
まず、実際の診断に先立ち、専門医によって、臨床所見と種々の画像検査から得られる情報をもとに、各膵嚢胞性疾患者の中から、特定の膵嚢胞性疾患の患者データが抽出された後、各膵嚢胞性疾患者毎に患者データが整理されて、各患者毎に下記に示すような入力項目に対応するデータが整理され、図５に示す入力データ表２０が作成される。
（１）年齢、性別、
（２）嚢胞の形状、個数、壁厚さ、充実成分、石灰化とその部位、
（３）造影の状況、
（４）嚢胞間信号強度、主膵管との交通、
（５）主膵管の拡張状況
この際、各入力項目に対応するデータとして、“0”、“1”、“2”の中から択一選択できるようにし、専門医以外の診断医でも、客観的に判断でき、簡単に入力できる内容にされる。 <Learning action>
First, prior to the actual diagnosis, patient data on specific pancreatic cystic disease was extracted from each pancreatic cystic disease patient by a specialist based on information obtained from clinical findings and various imaging tests. Thereafter, patient data is organized for each patient with pancreatic cystic disease, data corresponding to the input items shown below is organized for each patient, and an input data table 20 shown in FIG. 5 is created.
(1) Age, gender,
(2) Cys shape, number, wall thickness, solid component, calcification and its parts,
(3) Imaging conditions,
(4) Signal intensity between cysts, traffic with the main pancreatic duct,
(5) Expansion of the main pancreatic duct At this time, as data corresponding to each input item, one of “0”, “1”, and “2” can be selected. The content is easy to input.
まず、実際の診断に先立ち、専門医によって、臨床所見と種々の画像検査から得られる情報をもとに、各膵嚢胞性疾患者の中から、特定の膵嚢胞性疾患の患者データが抽出された後、各膵嚢胞性疾患者毎に患者データが整理されて、各患者毎に下記に示すような入力項目に対応するデータが整理され、図５に示す入力データ表２０が作成される。
（１）年齢、性別、
（２）嚢胞の形状、個数、壁厚さ、充実成分、石灰化とその部位、
（３）造影の状況、
（４）嚢胞間信号強度、主膵管との交通、
（５）主膵管の拡張状況
この際、各入力項目に対応するデータとして、“0”、“1”、“2”の中から択一選択できるようにし、専門医以外の診断医でも、客観的に判断でき、簡単に入力できる内容にされる。 <Learning action>
First, prior to the actual diagnosis, patient data on specific pancreatic cystic disease was extracted from each pancreatic cystic disease patient by a specialist based on information obtained from clinical findings and various imaging tests. Thereafter, patient data is organized for each patient with pancreatic cystic disease, data corresponding to the input items shown below is organized for each patient, and an input data table 20 shown in FIG. 5 is created.
(1) Age, gender,
(2) Cys shape, number, wall thickness, solid component, calcification and its parts,
(3) Imaging conditions,
(4) Signal intensity between cysts, traffic with the main pancreatic duct,
(5) Expansion of the main pancreatic duct At this time, as data corresponding to each input item, one of “0”, “1”, and “2” can be selected. The content is easy to input.
さらに、各膵嚢胞性疾患者毎に患者データが整理されて、各患者がどの膵嚢胞性疾患名、例えば下記に示すような膵嚢胞性疾患名のいずれに該当するか診断され、図６に示すような対応表２１が作成される。
（１）膵管内乳頭腫瘍ＩＰＭＮ、
（２）粘液性嚢胞腫瘍ＭＣＮ、
（３）漿液性嚢胞腺腫ＳＣＴ、
（４）Solid-pseudopapillary tumor ＳＰＴ、
（５）膵島腫瘍、
（６）膵腺房細胞腫瘍、
（７）膵仮性嚢胞、
（８）単純嚢胞、
（９）リンパ腫
この後、各入力項目データが入力層１７の各入力信号“Xi”に各々、割り当てるとともに、各膵嚢胞性疾患名が出力層１９の各出力信号“ym”に各々、割り当てられる。 Further, patient data is organized for each pancreatic cystic disease person, and it is diagnosed which pancreatic cystic disease name each patient corresponds to, for example, the name of the pancreatic cystic disease as shown below. A correspondence table 21 as shown is created.
(1) Intraductal papillary tumor IPMN,
(2) Myxoid cyst tumor MCN,
(3) Serous cystadenoma SCT,
(4) Solid-pseudopapillary tumor SPT,
(5) islet tumor,
(6) Pancreatic acinar cell tumor,
(7) Pancreatic pseudocyst,
(8) simple cyst,
(9) Lymphoma Thereafter, each input item data is assigned to each input signal “Xi” of theinput layer 17, and each pancreatic cystic disease name is assigned to each output signal “ym” of the output layer 19. .
（１）膵管内乳頭腫瘍ＩＰＭＮ、
（２）粘液性嚢胞腫瘍ＭＣＮ、
（３）漿液性嚢胞腺腫ＳＣＴ、
（４）Solid-pseudopapillary tumor ＳＰＴ、
（５）膵島腫瘍、
（６）膵腺房細胞腫瘍、
（７）膵仮性嚢胞、
（８）単純嚢胞、
（９）リンパ腫
この後、各入力項目データが入力層１７の各入力信号“Xi”に各々、割り当てるとともに、各膵嚢胞性疾患名が出力層１９の各出力信号“ym”に各々、割り当てられる。 Further, patient data is organized for each pancreatic cystic disease person, and it is diagnosed which pancreatic cystic disease name each patient corresponds to, for example, the name of the pancreatic cystic disease as shown below. A correspondence table 21 as shown is created.
(1) Intraductal papillary tumor IPMN,
(2) Myxoid cyst tumor MCN,
(3) Serous cystadenoma SCT,
(4) Solid-pseudopapillary tumor SPT,
(5) islet tumor,
(6) Pancreatic acinar cell tumor,
(7) Pancreatic pseudocyst,
(8) simple cyst,
(9) Lymphoma Thereafter, each input item data is assigned to each input signal “Xi” of the
そして、１枚目の入力データ表２０と、対応表２１が参照されて、１枚目の入力データ表２０の内容と、この入力データ表２０の番号に対応する膵嚢胞性疾患名が入力され、勾配法、最小二乗法などで、下記に示す如く各重み係数“Wji(t+1)”が最適化される。
Wji(t+1)=Wji(t)+C(tj-yj)*Xi …（３）
但し、C：学習係数
j：入力層１７、中間層１８、出力層１９を指定する番号
i：入力信号を指定する番号
t：学習番号
tj：教師信号
yj：出力信号 Then, referring to the first input data table 20 and the correspondence table 21, the contents of the first input data table 20 and the name of the pancreatic cystic disease corresponding to the number of the input data table 20 are input. Each weight coefficient “Wji (t + 1)” is optimized by the gradient method, the least square method, or the like as follows.
Wji (t + 1) = Wji (t) + C (tj-yj) * Xi (3)
Where C: learning coefficient j: number specifying theinput layer 17, intermediate layer 18, and output layer 19 i: number specifying the input signal t: learning number tj: teacher signal yj: output signal
Wji(t+1)=Wji(t)+C(tj-yj)*Xi …（３）
但し、C：学習係数
j：入力層１７、中間層１８、出力層１９を指定する番号
i：入力信号を指定する番号
t：学習番号
tj：教師信号
yj：出力信号 Then, referring to the first input data table 20 and the correspondence table 21, the contents of the first input data table 20 and the name of the pancreatic cystic disease corresponding to the number of the input data table 20 are input. Each weight coefficient “Wji (t + 1)” is optimized by the gradient method, the least square method, or the like as follows.
Wji (t + 1) = Wji (t) + C (tj-yj) * Xi (3)
Where C: learning coefficient j: number specifying the
次いで、２枚目の入力データ表２０と、対応表２１が参照されて、２枚目の入力データ表２０の内容と、この入力データ表２０の番号に対応する膵嚢胞性疾患名が入力され、勾配法、最小二乗法などで、（３）式に示す如く各重み係数“Wji(t+1)”が最適化される。
Next, referring to the second input data table 20 and the correspondence table 21, the contents of the second input data table 20 and the name of the pancreatic cystic disease corresponding to the number of the input data table 20 are input. Each weight coefficient “Wji (t + 1)” is optimized by the gradient method, the least square method, and the like as shown in the equation (3).
以下、残りの各入力データ表２０と、対応表２１が参照されて、残っている入力データ表２０の内容と、これらの入力データ表２０の番号に対応する各膵嚢胞性疾患名が、順次、入力され、勾配法、最小二乗法などで、（３）式に示す如く各重み係数“Wji(t+1)”が最適化される。
Hereinafter, the remaining input data table 20 and the correspondence table 21 are referred to, and the contents of the remaining input data table 20 and the names of pancreatic cystic diseases corresponding to the numbers of the input data table 20 are sequentially Each weight coefficient “Wji (t + 1)” is optimized by the gradient method, the least square method, etc. as shown in the equation (3).
これにより、最後には、特定の膵嚢胞性疾患名に対応する入力データ表２０の内容を入力したとき、特定の膵嚢胞性疾患名に対応する出力信号“y”が“１”になり、他の出力信号“y”が“０”になる。
Thus, finally, when the contents of the input data table 20 corresponding to the specific pancreatic cystic disease name are input, the output signal “y” corresponding to the specific pancreatic cystic disease name becomes “1”, The other output signal “y” becomes “0”.
そして、このように各重み係数“Wji(t+1)”が最適化された人工ニューラルネットワーク１５、すなわち学習済みの人工ニューラルネットワーク１５を持つコンピュータ５ａが病院側に引き渡される。
Then, the artificial neural network 15 in which each weight coefficient “Wji (t + 1)” is optimized as described above, that is, the computer 5a having the learned artificial neural network 15 is delivered to the hospital side.
なお、実際には、多層パーセプトロンでは、中間層１８の各重み係数“Wji(t+1)”の更新はなく、出力層１９の各重み係数“Wji(t+1)”の更新のみが行われる。
Actually, in the multilayer perceptron, each weighting factor “Wji (t + 1)” of the intermediate layer 18 is not updated, and only each weighting factor “Wji (t + 1)” of the output layer 19 is updated. Is called.
＜診断動作＞
また、病院では、臨床所見と種々の画像検査から得られる情報をもとに、診断医によって、図６に示すような入力項目に対応する入力データ表２０が作成される。 <Diagnosis operation>
In a hospital, an input data table 20 corresponding to input items as shown in FIG. 6 is created by a diagnostician based on information obtained from clinical findings and various image examinations.
また、病院では、臨床所見と種々の画像検査から得られる情報をもとに、診断医によって、図６に示すような入力項目に対応する入力データ表２０が作成される。 <Diagnosis operation>
In a hospital, an input data table 20 corresponding to input items as shown in FIG. 6 is created by a diagnostician based on information obtained from clinical findings and various image examinations.
この後、診断医によって、キーボード２、マウス３が操作されて、ディスプレイ４に表示される各入力項目毎に、３つある各チェックボックスの中から、入力データ表２０に対応する１つのチェックボックスが選択されて、“レ”点が入力される。
Thereafter, the check box corresponding to the input data table 20 is selected from the three check boxes for each input item displayed on the display 4 by operating the keyboard 2 and the mouse 3 by the diagnostician. Is selected and the “re” point is entered.
次いで、診断医によって、マウス３が操作されて、ディスプレイ４に表示されている診断開始ボタンがクリックされると、コンピュータ５ａによって、学習済みの人工ニューラルネットワーク（インストールされた膵嚢胞性疾患診断プログラムで記述された人工ニューラルネットワーク）１５が起動され、入力項目の内容に対応する膵嚢胞性疾患名が選択され、ディスプレイ４上に表示される。
Next, when the mouse 3 is operated by the diagnostician and the diagnosis start button displayed on the display 4 is clicked, the computer 5a causes the learned artificial neural network (with the installed pancreatic cystic disease diagnosis program). The described artificial neural network) 15 is activated, and the name of the pancreatic cystic disease corresponding to the content of the input item is selected and displayed on the display 4.
このように、この形態では、予め決められた数の入力項目を使用し、各入力項目毎に“0”、“1”、“2”の中から択一選択形式で、患者データが入力され、診断開始ボタンがクリックされたとき、予め学習させておいた人工ニューラルネットワーク１５を動作させて、患者データに対応する膵嚢胞性疾患名を表示させるようにしているので、学習処理期間を短縮させながら、臨床症状、所見と種々の通常の画像検査から得られる情報に基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができる（請求項１、２の効果）。
Thus, in this embodiment, a predetermined number of input items are used, and patient data is input in an alternative selection format from “0”, “1”, and “2” for each input item. When the diagnosis start button is clicked, the artificial neural network 15 trained in advance is operated to display the name of the pancreatic cystic disease corresponding to the patient data, so that the learning processing period is shortened. However, based on clinical symptoms, findings, and information obtained from various normal imaging tests, the patient's data is entered into each predetermined input item, and the name of the pancreatic cystic disease is automatically and accurately determined. (Effects of claims 1 and 2).
実際、標本群からいくつかの事例だけを抜き出してテスト事例とし、残りを訓練事例とするleave-some-out cross-validation（ＬＳＯＣＶ）法を使用し、予測精度を評価したところ、前述の入力項目から9種の疾患のうち、どの疾患であるかを予測することが可能であり、その全体での正診率は概ね“85%”と非常に高い予測効率であった。
In fact, we extracted only a few cases from the sample group and made them test cases, and the remaining-training cases were used to evaluate the prediction accuracy using the leave-some-out cross-validation (LSOCV) method. It was possible to predict which of the 9 diseases from the above, and the overall accuracy rate was approximately 85%, which was a very high prediction efficiency.
また、この形態では、ヘビサイド関数を使用するようにしているが、他の関数、例えばシグモイド関数などを使用しても良い。
In this embodiment, the snake side function is used, but other functions such as a sigmoid function may be used.
《第２形態》
図７は本発明による膵嚢胞性疾患診断装置のうち、請求項３に対応する一形態を示すブロック図である。なお、この図において、図１の各部と対応する部分には同じ符号が付してある。 << 2nd form >>
FIG. 7 is a block diagram showing an embodiment corresponding to claim 3 of the pancreatic cystic disease diagnosis apparatus according to the present invention. In this figure, parts corresponding to those in FIG. 1 are denoted by the same reference numerals.
図７は本発明による膵嚢胞性疾患診断装置のうち、請求項３に対応する一形態を示すブロック図である。なお、この図において、図１の各部と対応する部分には同じ符号が付してある。 << 2nd form >>
FIG. 7 is a block diagram showing an embodiment corresponding to claim 3 of the pancreatic cystic disease diagnosis apparatus according to the present invention. In this figure, parts corresponding to those in FIG. 1 are denoted by the same reference numerals.
この図に示す膵嚢胞性疾患診断装置１ｂが図１に示す膵嚢胞性疾患診断装置１ａと異なる点は、動径基底関数（Radial Basis Function：ＲＢＦ）型の人工ニューラルネットワーク（ＲＢＦネットワーク）３１（図８参照）を構築させる膵嚢胞性疾患診断プログラムを作成し、これをハードディスク機構１４ｂのプログラムエリア１３ｂに格納するようにしたことである。
The pancreatic cystic disease diagnostic apparatus 1b shown in this figure is different from the pancreatic cystic disease diagnostic apparatus 1a shown in FIG. 1 in that a radial basis function (RBF) type artificial neural network (RBF network) 31 ( 8), a pancreatic cystic disease diagnosis program is created and stored in the program area 13b of the hard disk mechanism 14b.
ＲＢＦネットワーク３１は、図８に示す如く複数のユニット３２によって構成され、予め指定された入力項目の内容に対応する複数の入力信号を各々取り込み、複数の出力信号を出力する入力層３３と、複数のユニット３２によって構成され、入力層３３を構成する各ユニット３２の出力信号を取り込み、複数の出力信号を出力する中間層３４と、複数のユニット３２によって構成され、中間層３４を構成する各ユニット３２の出力信号を取り込み、複数の出力信号を出力する出力層３５とを備えており、膵嚢胞性疾患を診断を開始する前に、多数の臨床データが使用されて、学習処理が行われて、各ユニット３２の重み係数“Wij”の値が最適化される。そして、診断時には、キーボード２、マウス３が操作されて、ディスプレイ４上に表示されている各入力項目に患者の検査結果が入力され、診断開始ボタンがクリックされたとき、入力層３３を構成する各ユニット３２から検査結果に対応した出力信号が出力されて、中間層３４を構成する各ユニット３２で処理され、出力層３５を構成する各ユニット３２から診断結果に対応した出力信号が出力される。
As shown in FIG. 8, the RBF network 31 includes a plurality of units 32, each of which receives a plurality of input signals corresponding to the contents of input items specified in advance, and outputs a plurality of output signals. The unit 32 includes the intermediate layer 34 that takes in the output signal of each unit 32 that constitutes the input layer 33 and outputs a plurality of output signals, and each unit that comprises the plurality of units 32 and constitutes the intermediate layer 34. An output layer 35 that captures 32 output signals and outputs a plurality of output signals. Before starting diagnosis of pancreatic cystic disease, a large amount of clinical data is used to perform a learning process. The value of the weight coefficient “Wij” of each unit 32 is optimized. At the time of diagnosis, the keyboard 2 and the mouse 3 are operated, the patient test results are input to each input item displayed on the display 4, and the input layer 33 is configured when the diagnosis start button is clicked. An output signal corresponding to the test result is output from each unit 32, processed by each unit 32 configuring the intermediate layer 34, and an output signal corresponding to the diagnosis result is output from each unit 32 configuring the output layer 35. .
各ユニット３２は、図９に示すように複数の入力端子と１つの出力端子とを持ち、各入力信号“x”に対応する動径基底関数、例えば下記に示す（４）式で表され、図１０に示すように中心で最大値（あるいは最小値）を取り、そこから離れるにしたがって単調に減少（あるいは増加）していくようなガウス関数演算を行って、各動径基底関数“φi(x)”を求めるとともに、下記に示す（５）式を使用して、これらの各動径基底関数“φi(x)”と、各重み係数“Wi”との乗算値“Wi*φi(x)”を求め、さらに各乗算値“Wi*φi(x)”の総和“ΣWi*φi(x)”に予め設定されているバイアス値“w0”を加算して、出力関数“y”を求め、これを出力信号として出力する。
φi(x) = (x - ci)T (x - ci) / (2 σ2i) …（４）
但し、φi(x)：ｉ番目の動径基底関数
i：入力信号、動径基底関数の番号を示す値（i=1,2,...,N）
ci：ｉ番目の動径基底関数（ガウス関数）の中心
T：ベクトルを転置させる記号
σ：ガウス関数の標準偏差で、幅を決めるパラメータ
x：入力信号
x - ci：入力信号とガウス関数中心とのユークリッド距離（ベクトル表記）
但し、y：出力信号
i：入力信号、動径基底関数の番号を示す値（i=1,2,...,N）
φi(x)：i番目の入力信号“x”に対応する動径基底関数
w0：バイアスの値（y方向へのずれ）
wi：i番目の動径基底関数に対する重み係数 Eachunit 32 has a plurality of input terminals and one output terminal as shown in FIG. 9, and is represented by a radial basis function corresponding to each input signal “x”, for example, the following equation (4): As shown in FIG. 10, a Gaussian function operation is performed in which a maximum value (or minimum value) is taken at the center and monotonously decreasing (or increasing) as the distance from the center increases, and each radial basis function “φi ( x) ”and using the following equation (5), the product of each radial basis function“ φi (x) ”and each weight coefficient“ Wi ”is multiplied by“ Wi * φi (x ) ”And add the preset bias value“ w0 ”to the sum“ ΣWi * φi (x) ”of each multiplication value“ Wi * φi (x) ”to obtain the output function“ y ”. This is output as an output signal.
φi (x) = (x-ci) T (x-ci) / (2 σ2i) (4)
Where φi (x): i-th radial basis function i: value indicating input signal, radial basis function number (i = 1, 2,..., N)
ci: Center of i-th radial basis function (Gaussian function) T: Symbol for transposing the vector σ: Standard deviation of Gaussian function, parameter that determines width x: Input signal x-ci: Input signal and Gaussian function center Euclidean distance (vector notation)
Where y: output signal i: input signal, value indicating radial basis function number (i = 1,2, ..., N)
φi (x): Radial basis function corresponding to the i-th input signal “x” w0: Bias value (shift in y direction)
wi: Weighting factor for the i-th radial basis function
φi(x) = (x - ci)T (x - ci) / (2 σ2i) …（４）
但し、φi(x)：ｉ番目の動径基底関数
i：入力信号、動径基底関数の番号を示す値（i=1,2,...,N）
ci：ｉ番目の動径基底関数（ガウス関数）の中心
T：ベクトルを転置させる記号
σ：ガウス関数の標準偏差で、幅を決めるパラメータ
x：入力信号
x - ci：入力信号とガウス関数中心とのユークリッド距離（ベクトル表記）
i：入力信号、動径基底関数の番号を示す値（i=1,2,...,N）
φi(x)：i番目の入力信号“x”に対応する動径基底関数
w0：バイアスの値（y方向へのずれ）
wi：i番目の動径基底関数に対する重み係数 Each
φi (x) = (x-ci) T (x-ci) / (2 σ2i) (4)
Where φi (x): i-th radial basis function i: value indicating input signal, radial basis function number (i = 1, 2,..., N)
ci: Center of i-th radial basis function (Gaussian function) T: Symbol for transposing the vector σ: Standard deviation of Gaussian function, parameter that determines width x: Input signal x-ci: Input signal and Gaussian function center Euclidean distance (vector notation)
φi (x): Radial basis function corresponding to the i-th input signal “x” w0: Bias value (shift in y direction)
wi: Weighting factor for the i-th radial basis function
そして、このように構成されるＲＢＦネットワーク３１に対し、上述した学習動作を行わせることにより、上述した形態と同様に、図５に示す入力項目毎のデータと、図６に示す診断結果とが非線形である場合にも、臨床症状、所見と種々の通常の画像検査から得られる情報に基づき、予め決められた各入力項目に患者データを入力するだけで、膵嚢胞性疾患名を自動的に、かつ正確に判定することができる（請求項３の効果）。
Then, by causing the RBF network 31 configured as described above to perform the learning operation described above, the data for each input item illustrated in FIG. 5 and the diagnosis result illustrated in FIG. Even if it is non-linear, the patient's pancreatic cystic disease name is automatically entered by simply entering patient data into each predetermined entry based on information obtained from clinical symptoms, findings and various normal imaging tests. And can be determined accurately (effect of claim 3).
《第３形態》
図１１は本発明による膵嚢胞性疾患診断装置のうち、請求項４に対応する一形態を示すブロック図である。なお、この図において、図１の各部と対応する部分には同じ符号が付してある。 << 3rd form >>
FIG. 11 is a block diagram showing an embodiment corresponding to claim 4 of the pancreatic cystic disease diagnosis apparatus according to the present invention. In this figure, parts corresponding to those in FIG. 1 are denoted by the same reference numerals.
図１１は本発明による膵嚢胞性疾患診断装置のうち、請求項４に対応する一形態を示すブロック図である。なお、この図において、図１の各部と対応する部分には同じ符号が付してある。 << 3rd form >>
FIG. 11 is a block diagram showing an embodiment corresponding to claim 4 of the pancreatic cystic disease diagnosis apparatus according to the present invention. In this figure, parts corresponding to those in FIG. 1 are denoted by the same reference numerals.
この図に示す膵嚢胞性疾患診断装置１ｃが図１に示す膵嚢胞性疾患診断装置１ａと異なる点は、コンピュータ５ｃ内に通信回路４１を設け、ハブ（図示は省略する）を介して、通信回路４１とＬＡＮケーブル（イーサネットケーブル、または無線ＬＡＮ回線）４２とを接続させるようにしたことである。
The pancreatic cystic disease diagnostic apparatus 1c shown in this figure is different from the pancreatic cystic disease diagnostic apparatus 1a shown in FIG. 1 in that a communication circuit 41 is provided in the computer 5c, and communication is performed via a hub (not shown). That is, the circuit 41 and a LAN cable (Ethernet cable or wireless LAN line) 42 are connected.
そして、ＬＡＮケーブル４２の各ハブ（図示は省略する）に接続されたＵＳ（Ultrasonography：超音波断層撮影）装置４３、ＣＴ（Computer
Tomography：コンピューター断層）装置４４、ＭＲＩ（Magnetic Resonance Imaging：磁気共鳴画像）装置４５、ＥＵＳ（endoscopic ultrasonography：内視鏡的超音波断層）装置４６、ＥＲＣＰ（endoscopic retrograde cholangio-pancreatography：内視鏡的逆行性膵胆管造影）装置４７などから直接、診断結果を取り込み、ディスプレイ４に表示するとともに、キーボード２、マウス３が操作されて、診断結果解析指示が入力されたとき、プログラムエリア１３ｃに格納された診断結果解析プログラムを動作させて、ＵＳ装置４３の診断結果～ＥＲＣＰ装置４７の診断結果を“０”、“１”、“２”に量子化させた後、各入力項目のデータとして、処理するようにしたりするようにしたことである。 A US (Ultrasonography)apparatus 43 connected to each hub (not shown) of the LAN cable 42, CT (Computer)
Tomography: computer tomography (44), MRI (magnetic resonance imaging) 45, EUS (endoscopic ultrasonography) 46, ERCP (endoscopic retrograde cholangio-pancreatography) (Diagnostic pancreatobiliography) The diagnosis result is directly captured from theapparatus 47 and displayed on the display 4, and when the keyboard 2 and mouse 3 are operated to input the analysis result analysis instruction, the diagnosis result is stored in the program area 13c. Execute the diagnostic result analysis program to quantize the diagnostic result of the US device 43 to the diagnostic result of the ERCP device 47 into “0”, “1”, “2”, and then process it as data of each input item It is to do so.
Tomography：コンピューター断層）装置４４、ＭＲＩ（Magnetic Resonance Imaging：磁気共鳴画像）装置４５、ＥＵＳ（endoscopic ultrasonography：内視鏡的超音波断層）装置４６、ＥＲＣＰ（endoscopic retrograde cholangio-pancreatography：内視鏡的逆行性膵胆管造影）装置４７などから直接、診断結果を取り込み、ディスプレイ４に表示するとともに、キーボード２、マウス３が操作されて、診断結果解析指示が入力されたとき、プログラムエリア１３ｃに格納された診断結果解析プログラムを動作させて、ＵＳ装置４３の診断結果～ＥＲＣＰ装置４７の診断結果を“０”、“１”、“２”に量子化させた後、各入力項目のデータとして、処理するようにしたりするようにしたことである。 A US (Ultrasonography)
Tomography: computer tomography (44), MRI (magnetic resonance imaging) 45, EUS (endoscopic ultrasonography) 46, ERCP (endoscopic retrograde cholangio-pancreatography) (Diagnostic pancreatobiliography) The diagnosis result is directly captured from the
このようにすることにより、入力項目毎のデータを入力する際、各診断医の検査結果の入力内容がばらつかないようにすることができ、より正確な診断を行わせることができる（請求項４の効果）。
By doing so, when inputting data for each input item, it is possible to prevent variations in the input contents of the examination results of each diagnostician, and it is possible to perform more accurate diagnosis. Effect of 4).
《第４形態》
図１２は本発明による膵嚢胞性疾患診断装置のうち、請求項５に対応する一形態を示すブロック図である。なお、この図において、図１１の各部と対応する部分には同じ符号が付してある。 << 4th form >>
FIG. 12 is a block diagram showing an embodiment corresponding to claim 5 of the pancreatic cystic disease diagnosis apparatus according to the present invention. In this figure, parts corresponding to those in FIG. 11 are given the same reference numerals.
図１２は本発明による膵嚢胞性疾患診断装置のうち、請求項５に対応する一形態を示すブロック図である。なお、この図において、図１１の各部と対応する部分には同じ符号が付してある。 << 4th form >>
FIG. 12 is a block diagram showing an embodiment corresponding to claim 5 of the pancreatic cystic disease diagnosis apparatus according to the present invention. In this figure, parts corresponding to those in FIG. 11 are given the same reference numerals.
この図に示す膵嚢胞性疾患診断装置１ｄが図１に示す膵嚢胞性疾患診断装置１ｃと異なる点は、ＬＡＮケーブル４２とインターネット回線とを接続するゲートウェイ装置（または、ルータ装置）５１を設けるとともに、ハードディスク機構１４ｄのプログラムエリア１３ｄにリモートプログラムを格納し、インターネットに接続されたサーバ装置（図示は省略する）によって、モートプログラムを動作させて、膵嚢胞性疾患診断プログラムを更新させたり、膵嚢胞性疾患診断プログラムで記述されている人工ニューラルネットワーク１５などを学習させたりするようにしたことである。
The pancreatic cystic disease diagnostic apparatus 1d shown in this figure is different from the pancreatic cystic disease diagnostic apparatus 1c shown in FIG. 1 in that a gateway device (or router device) 51 for connecting the LAN cable 42 and the Internet line is provided. The remote program is stored in the program area 13d of the hard disk mechanism 14d, and the mote program is operated by a server device (not shown) connected to the Internet to update the pancreatic cystic disease diagnostic program, The artificial neural network 15 described in the sexually transmitted disease diagnosis program is learned.
このようにすることにより、サーバ装置によって、膵嚢胞性疾患診断プログラムを更新させたり、膵嚢胞性疾患診断プログラムで記述されている人工ニューラルネットワーク１５（あるいは、ＲＢＦネットワーク３１）を学習させて、膵嚢胞性疾患診断精度をさらに向上させることができる（請求項５の効果）。
By doing so, the server device updates the pancreatic cystic disease diagnosis program or learns the artificial neural network 15 (or RBF network 31) described in the pancreatic cystic disease diagnosis program, Cystic disease diagnosis accuracy can be further improved (effect of claim 5).
《他の形態》
また、上述した説明から明らかなように、本発明による膵嚢胞性疾患診断装置１ａ～１ｄでは、人工ニューラルネットワーク１５、ＲＢＦネットワーク３１を使用した診断を行っているので、簡便な入力項目に患者のデータを入力するだけで、高い診断率で、膵嚢胞性疾患名を判定させることができる。今後、症例数の蓄積とより、詳細な入力項目を設定することにより、さらに予測効率があがると考えられる。 《Other forms》
Further, as is clear from the above description, in the pancreatic cystic disease diagnosis apparatus 1a to 1d according to the present invention, the diagnosis is performed using the artificialneural network 15 and the RBF network 31, so that the patient's simple input items are included. By simply inputting data, the name of the pancreatic cystic disease can be determined with a high diagnosis rate. In the future, it is considered that the prediction efficiency can be further improved by accumulating the number of cases and setting more detailed input items.
また、上述した説明から明らかなように、本発明による膵嚢胞性疾患診断装置１ａ～１ｄでは、人工ニューラルネットワーク１５、ＲＢＦネットワーク３１を使用した診断を行っているので、簡便な入力項目に患者のデータを入力するだけで、高い診断率で、膵嚢胞性疾患名を判定させることができる。今後、症例数の蓄積とより、詳細な入力項目を設定することにより、さらに予測効率があがると考えられる。 《Other forms》
Further, as is clear from the above description, in the pancreatic cystic disease diagnosis apparatus 1a to 1d according to the present invention, the diagnosis is performed using the artificial
また、このような人工ニューラルネットワーク１５、ＲＢＦネットワーク３１を使用した診断方法は、膵嚢胞性疾患の画像診断だけではなく、一般疾患診断へ応用可能であり新たな診断モダリティーとなりえると考えられる。
Also, such a diagnostic method using the artificial neural network 15 and the RBF network 31 can be applied not only to image diagnosis of pancreatic cystic disease but also to general disease diagnosis, and can be a new diagnostic modality.
本発明は、医療機関などで使用される膵嚢胞性疾患診断装置に係わり、特に人工ニューラルネットワークを使用して、膵嚢胞性疾患を診断する膵嚢胞性疾患診断装置に関し、産業上の利用可能性を有する。
The present invention relates to a pancreatic cystic disease diagnostic apparatus used in medical institutions, and more particularly to a pancreatic cystic disease diagnostic apparatus that diagnoses pancreatic cystic disease using an artificial neural network, and has industrial applicability. Have
１ａ～１ｄ：膵嚢胞性疾患診断装置
２：キーボード
３：マウス
４：ディスプレイ
５ａ～５ｄ：コンピュータ
６：システムバス
７：キーボードインタフェース回路
８：マウスインタフェース回路
９：ディスプレイインタフェース回路
１０：ＣＰＵ回路
１１：メモリ回路
１２：ＯＳエリア
１３ａ～１３ｄ：プログラムエリア
１４ａ～１４ｄ：ハードディスク機構
１５：人工ニューラルネットワーク
１６：ニューロン
１７：入力層
１８：中間層
１９：出力層
２０：入力データ表
２１：対応表
３１：ＲＢＦネットワーク
３２：ユニット
３３：入力層
３４：中間層
３５：出力層
４１：通信回路４１
４２：ＬＡＮケーブル
４３：ＵＳ装置
４４：ＣＴ装置
４５：ＭＲＩ装置
４６：ＥＵＳ装置
４７：ＥＲＣＰ装置
５１：ゲートウェイ装置 1a to 1d: Pancreatic cystic disease diagnosis device 2: Keyboard 3: Mouse 4:Display 5a to 5d: Computer 6: System bus 7: Keyboard interface circuit 8: Mouse interface circuit 9: Display interface circuit 10: CPU circuit 11: Memory Circuit 12: OS area 13a to 13d: Program area 14a to 14d: Hard disk mechanism 15: Artificial neural network 16: Neuron 17: Input layer 18: Intermediate layer 19: Output layer 20: Input data table 21: Correspondence table 31: RBF network 32: Unit 33: Input layer 34: Intermediate layer 35: Output layer 41: Communication circuit 41
42: LAN cable 43: US device 44: CT device 45: MRI device 46: EUS device 47: ERCP device 51: Gateway device
２：キーボード
３：マウス
４：ディスプレイ
５ａ～５ｄ：コンピュータ
６：システムバス
７：キーボードインタフェース回路
８：マウスインタフェース回路
９：ディスプレイインタフェース回路
１０：ＣＰＵ回路
１１：メモリ回路
１２：ＯＳエリア
１３ａ～１３ｄ：プログラムエリア
１４ａ～１４ｄ：ハードディスク機構
１５：人工ニューラルネットワーク
１６：ニューロン
１７：入力層
１８：中間層
１９：出力層
２０：入力データ表
２１：対応表
３１：ＲＢＦネットワーク
３２：ユニット
３３：入力層
３４：中間層
３５：出力層
４１：通信回路４１
４２：ＬＡＮケーブル
４３：ＵＳ装置
４４：ＣＴ装置
４５：ＭＲＩ装置
４６：ＥＵＳ装置
４７：ＥＲＣＰ装置
５１：ゲートウェイ装置 1a to 1d: Pancreatic cystic disease diagnosis device 2: Keyboard 3: Mouse 4:
42: LAN cable 43: US device 44: CT device 45: MRI device 46: EUS device 47: ERCP device 51: Gateway device
Claims (5)
- 臨床症状、所見、各画像検査のいずれか１つ以上の入力データを用いて、膵嚢胞性疾患名を判定する膵嚢胞性疾患診断装置において、
人工ニューラルネットワークに、各入力項目の内容と、各疾患名とを学習させ、各入力項目の内容が入力されたとき、前記人工ニューラルネットワークを動作させて、疾患名を出力する、ことを特徴とする膵嚢胞性疾患診断装置。 In the diagnostic apparatus for pancreatic cystic disease that determines the name of pancreatic cystic disease using any one or more input data of clinical symptoms, findings, and image examinations,
The artificial neural network is made to learn the contents of each input item and each disease name, and when the contents of each input item are input, the artificial neural network is operated to output the disease name. Pancreatic cystic disease diagnostic device. - 請求項１に記載の膵嚢胞性疾患診断装置において、
前記人工ニューラルネットワークを構成するユニットとして、ヘビサイド関数、またはシグモイド関数を使用したニューロンを使用する、ことを特徴とする膵嚢胞性疾患診断装置。 In the pancreatic cystic disease diagnostic apparatus according to claim 1,
A device for diagnosing pancreatic cystic disease, wherein a neuron using a snake side function or a sigmoid function is used as a unit constituting the artificial neural network. - 請求項１に記載の膵嚢胞性疾患診断装置において、
前記人工ニューラルネットワークを構成するユニットとして、動径基底関数を使用したニューロンを使用する、ことを特徴とする膵嚢胞性疾患診断装置。 In the pancreatic cystic disease diagnostic apparatus according to claim 1,
A device for diagnosing pancreatic cystic disease, wherein a neuron using a radial basis function is used as a unit constituting the artificial neural network. - 請求項１、２、３のいずれかに記載の膵嚢胞性疾患診断装置において、
有線ケーブル、または無線回線を介して、ＵＳ装置、ＣＴ装置、ＭＲＩ装置、ＥＵＳ装置、ＥＲＣＰ装置から、直接、診断結果を取り込む、ことを特徴とする膵嚢胞性疾患診断装置。 In the diagnostic apparatus for pancreatic cystic disease according to any one of claims 1, 2, and 3,
A diagnostic apparatus for pancreatic cystic disease, which directly captures a diagnostic result from a US apparatus, CT apparatus, MRI apparatus, EUS apparatus, or ERCP apparatus via a wired cable or a wireless line. - 請求項１、２、３、４のいずれかに記載の膵嚢胞性疾患診断装置において、
有線ケーブル、無線回線、インターネット回線のいずれかを介して、膵嚢胞性疾患診断プログラムが更新される処理、膵嚢胞性疾患診断プログラムで記述されている人工ニューラルネットワーク、ＲＢＦネットワークの学習処理が行われる、ことを特徴とする膵嚢胞性疾患診断装置。 In the diagnostic apparatus for pancreatic cystic disease according to any one of claims 1, 2, 3, and 4,
The process of updating the pancreatic cystic disease diagnosis program, the artificial neural network described in the pancreatic cystic disease diagnosis program, and the learning process of the RBF network are performed via any of a wired cable, a wireless line, and an Internet line. Diagnostic apparatus for pancreatic cystic disease characterized by the above.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2009047733A JP2010200881A (en) | 2009-03-02 | 2009-03-02 | Pancreas cystic disease diagnostic apparatus |
| JP2009-047733 | 2009-03-02 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2010101150A1 true WO2010101150A1 (en) | 2010-09-10 |
Family
ID=42709706
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2010/053346 WO2010101150A1 (en) | 2009-03-02 | 2010-03-02 | Device for diagnosing pancreatic cystic disease |
Country Status (2)
| Country | Link |
|---|---|
| JP (1) | JP2010200881A (en) |
| WO (1) | WO2010101150A1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112889116A (en) * | 2018-10-05 | 2021-06-01 | 第一百欧有限公司 | System and method for searching pathological image |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5484309B2 (en) * | 2010-12-20 | 2014-05-07 | 紀文 日比 | Prognosis prediction method for patients with ulcerative colitis undergoing CAP |
| KR101659578B1 (en) | 2015-09-01 | 2016-09-23 | 삼성전자주식회사 | Method and apparatus for processing magnetic resonance imaging |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0512351A (en) * | 1991-07-02 | 1993-01-22 | Toshiba Corp | Diagnosis assistance system |
| JPH05277119A (en) * | 1992-03-31 | 1993-10-26 | Nuclear Fuel Ind Ltd | Cancer diagnostic device |
| JPH0824227A (en) * | 1994-07-19 | 1996-01-30 | Hitachi Medical Corp | Medical image diagnosing apparatus |
| JP2006502481A (en) * | 2002-10-03 | 2006-01-19 | スコット・ラボラトリーズ・インコーポレイテッド | Neural network in sedation and analgesia system |
| JP2006043007A (en) * | 2004-08-02 | 2006-02-16 | Fujitsu Ltd | Diagnosis support program and diagnosis support apparatus |
-
2009
- 2009-03-02 JP JP2009047733A patent/JP2010200881A/en active Pending
-
2010
- 2010-03-02 WO PCT/JP2010/053346 patent/WO2010101150A1/en active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0512351A (en) * | 1991-07-02 | 1993-01-22 | Toshiba Corp | Diagnosis assistance system |
| JPH05277119A (en) * | 1992-03-31 | 1993-10-26 | Nuclear Fuel Ind Ltd | Cancer diagnostic device |
| JPH0824227A (en) * | 1994-07-19 | 1996-01-30 | Hitachi Medical Corp | Medical image diagnosing apparatus |
| JP2006502481A (en) * | 2002-10-03 | 2006-01-19 | スコット・ラボラトリーズ・インコーポレイテッド | Neural network in sedation and analgesia system |
| JP2006043007A (en) * | 2004-08-02 | 2006-02-16 | Fujitsu Ltd | Diagnosis support program and diagnosis support apparatus |
Non-Patent Citations (1)
| Title |
|---|
| MASANORI KOJIMA: "EPCP Shiko Rei kara Mita MRCP to ERCP no Tekio", PROGRESS OF DIGESTIVE ENDOSCOPY, vol. 68, no. 2, 10 June 2006 (2006-06-10), pages 73 - 76 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112889116A (en) * | 2018-10-05 | 2021-06-01 | 第一百欧有限公司 | System and method for searching pathological image |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2010200881A (en) | 2010-09-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Alugubelli | Exploratory study of artificial intelligence in healthcare | |
| Freer et al. | Social frailty: the importance of social and environmental factors in predicting frailty in older adults | |
| JPWO2019150813A1 (en) | Data processing equipment and methods, recognition equipment, learning data storage equipment, machine learning equipment and programs | |
| WO2013181638A2 (en) | Image based analytical systems and processes | |
| JP2009533782A (en) | Personal prognostic modeling in medical planning | |
| TWI521467B (en) | Nursing decision support system | |
| US20200243200A1 (en) | System and method for determining veracity of patient diagnoses within one or more electronic health records | |
| US20230154575A1 (en) | Systems and Methods for Mental Health Care Delivery Via Artificial Intelligence | |
| Young et al. | Stress testing reveals gaps in clinic readiness of image-based diagnostic artificial intelligence models | |
| Kurz et al. | Uncertainty estimation in medical image classification: systematic review | |
| Long et al. | Coping strategies and social support among caregivers of patients with cancer: a cross-sectional study in Vietnam | |
| TW202217843A (en) | Method and computer program product and apparatus for remotely diagnosing tongues based on deep learning | |
| Mridha et al. | Automated stroke prediction using machine learning: An explainable and exploratory study with a web application for early intervention | |
| Mese et al. | Improving radiology workflow using ChatGPT and artificial intelligence | |
| WO2010101150A1 (en) | Device for diagnosing pancreatic cystic disease | |
| Kyrimi et al. | Bayesian Networks in Healthcare: the chasm between research enthusiasm and clinical adoption | |
| Sharma et al. | An explainable artificial intelligence based prospective framework for COVID-19 risk prediction | |
| US20210151188A1 (en) | Eating Disorder Diagnosis | |
| JP7238705B2 (en) | Medical care support method, medical care support system, learning model generation method, and medical care support program | |
| CN112397195B (en) | Method, apparatus, electronic device and medium for generating physical examination model | |
| JP2022059448A (en) | Diagnosis and treatment support system | |
| McLeod-Sordjan et al. | Effectiveness of patient-caregiver dyad discharge interventions on hospital readmissions of elderly patients with community acquired pneumonia: a systematic review | |
| JP4871202B2 (en) | Medical education support system and medical education support method | |
| JP2014199597A (en) | Medical examination support program, medical examination support method and medical examination support device | |
| Megalla et al. | Using the 4M model to screen geriatric patients in the emergency department |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 10748745 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 10748745 Country of ref document: EP Kind code of ref document: A1 |