CN110349662B - Cross-image set outlier sample discovery method and system for filtering lung mass misdetection results - Google Patents
Cross-image set outlier sample discovery method and system for filtering lung mass misdetection results Download PDFInfo
- Publication number
- CN110349662B CN110349662B CN201910436092.0A CN201910436092A CN110349662B CN 110349662 B CN110349662 B CN 110349662B CN 201910436092 A CN201910436092 A CN 201910436092A CN 110349662 B CN110349662 B CN 110349662B
- Authority
- CN
- China
- Prior art keywords
- image
- suspected
- points
- region
- lung
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Radiology & Medical Imaging (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Pathology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of data mining, pattern recognition and image processing, and particularly relates to a cross-image set outlier sample finding method and system for filtering lung mass misdetection results. The outlier sample discovery method of the invention comprises the following steps: scanning a human body through medical imaging equipment to obtain a CT image sequence arranged in sequence; marking the area occupied by the tumor at the corresponding coordinate position of the corresponding image frame where the suspected lung tumor is located; extracting image features for each suspected mass; and analyzing the image characteristics of all suspected tumor areas of all patients by adopting an outlier sample detection technology, and filtering the suspected tumor areas in the image set corresponding to the feature space outlier sample. The outlier of each sample corresponds to false detection, and the number of false detection samples is reduced by filtering samples with large outliers, so that false alarms are reduced.
Description
Technical Field
The invention belongs to the technical field of data mining, pattern recognition and image processing, and particularly relates to a cross-image set outlier sample finding method and system for filtering lung mass misdetection results.
Background
Computer-aided diagnosis of medical images has been of great interest. Various medical image analysis and computer-aided diagnosis methods are developed, however, various methods suffer from false detection. The misdetection results bring troubles to doctors, and even lose the significance of auxiliary diagnosis.
How to further judge the tumor detection result in the medical image through a post-processing technology is a problem which needs to be solved urgently, the previous research is to further filter abnormal samples in the detection result based on morphological analysis of the detection result of a single image, but the information of the single image is limited, the abnormal samples have various morphologies and are difficult to comprehensively summarize, and the knowledge depended on the abnormal detection is incomplete.
A new solution has recently been proposed: knowledge is found from the entire set of images for anomaly detection [1]. However, the method of document [1] is also a method of supervised machine learning, which requires a large number of expensive image labels, and such large-scale labels are often not easily available.
The invention provides an abnormal sample finding method based on unsupervised machine learning, which is used for filtering false detection results in computer-aided lung lump detection to reduce false alarms. Firstly, traversing the tumor detection results of CT images of all patients, and extracting the characteristics of all tumors; observation shows that the samples (normal samples) corresponding to the correct detection results present aggregate situations in the feature space and have consistency in features, while the false detection samples present scattered random distribution in the feature space due to different features and are often far away from the normal samples. Therefore, the outlier sample detection technology in the data mining category is adopted, the outlier degree of each sample is analyzed in an unsupervised mode in the image feature space, and the number of false detection samples is reduced by filtering the outlier samples, so that false alarms are reduced.
Reference documents:
[1]Ke Yan,Xiaosong Wang,Le Lu,Ling Zhang,Adam P.Harrison,MohammadhadiBagheri, and Ronald M.Summers,“Deep Lesion Graphs in the Wild:Relationship Learning and Organization of Significant Radiology Image Findings in a Diverse Large-scale Lesion Database”,2018IEEE/CVF Conference on Computer Vision and Pattern Recognition,pp. 9261-9270.。
disclosure of Invention
The invention aims to provide a cross-image set outlier sample finding method and a cross-image set outlier sample finding system for filtering lung mass misdetection results, which are used for reducing false alarms of computer-aided diagnosis.
The invention provides a cross-image-set outlier sample finding method for filtering a lung mass misdetection result, which is used for detecting outliers in a characteristic space of an area judged to be a suspected mass in the early stage by traversing the whole image set, and comprises the following specific steps of:
(a) Scanning a human body through medical imaging equipment to obtain a CT image sequence which is arranged in sequence, wherein two slices which are adjacent in space correspond to images of adjacent frames;
(b) Marking the area occupied by the tumor at the corresponding coordinate position of the corresponding image frame where the suspected lung tumor is located;
(c) Extracting image features for each suspected mass;
further, the image feature extraction of the suspected mass comprises:
calculating the mean value of CT values of the suspected lump area as the characteristic of the area;
calculating the standard deviation of the CT value of the suspected lump area as the characteristic of the area;
(d) Based on an outlier detection (outlier detection) technology, the image characteristics of all suspected tumor areas of all patients are analyzed, and the suspected tumor areas in the image set corresponding to the feature space outlier samples are filtered out.
Further, there may be specifically 2 representative calculation methods as follows:
the first calculation method comprises the following steps:
(1) Pretreatment: assuming that there are L suspected masses, the feature corresponding to each mass is a vector, and the feature space has L samples { f } 1 ,f 2 ,…,f L Normalizing for each characteristic dimension such that the values of the L samples in the characteristic dimension obey a standard normal distribution;
(2) Judging an outlier sample by adopting a K nearest neighbor method: sample f i Average distance O to its K nearest neighbors i As an abnormality metric index (i =1,2, …, L) { O 1 ,O 2 ,…,O L Sorting from large to small, and taking the corresponding first L' samples as outlier samples, where O L′ Is { O 1 ,O 2 ,…,O L Where the inflection point appears after the sorting.
The second calculation method comprises the following steps:
(1) Pretreatment: assuming that there are L suspected masses, the feature corresponding to each mass is a vector, and the feature space has L samples { f } 1 ,f 2 ,…,f L Normalizing for each characteristic dimension such that the values of the L samples in the characteristic dimension obey a standard normal distribution;
(2) Establishing a connectivity graph between samples: calculating Euclidean distance between every two L samples:
{d ij |i＝1,2,…,L；j＝1,2,…,L},
to { d ij I =1,2, …, L; j =1,2, …, L } order, finding the value d at which the sequence inflection point occurs T And calculating:
in the above formula, i =1,2, …, L, j =1,2, …, L. If L samples are considered as L nodes constituting the graph model, r ij =1 denotes the communication between sample i and sample j, r ij And =0 indicates that the two are not connected. { r ij I =1,2, …, L; the connected relation among all samples recorded by j =1,2, …, L } forms a graph model, all connected subgraphs are found, and the samples contained in all subgraphs except the largest connected subgraph (the subgraph containing the largest number of samples) are taken as outlier samples.
Based on the above discovery method, the present invention also relates to a cross-image set outlier sample discovery system for filtering lung mass misdetection results. The system comprises 4 modules: the system comprises a human body image sequence acquisition and sorting module, a suspected lung lump region marking module, a suspected lump extraction image feature module and an outlier sample detection module; these 4 modules correspond in turn to the operational content of steps (a), (b), (c), (d) in the cross-image set outlier sample discovery method for filtering lung mass misdetection results.
The invention adopts the outlier sample detection technology in the data mining category, analyzes the outlier degree of each sample in an unsupervised mode in the image characteristic space, and reduces the number of false detection samples by filtering the outlier samples, thereby reducing false alarms.
Drawings
Figure 1 is a feature space sample distribution (circles are outlier samples found).
Figure 2 shows correctly detected masses (inside the grey box).
Fig. 3 shows the detected tumor in the CT image corresponding to the outlier sample.
Fig. 4 shows the detection result of tumor mass in CT image corresponding to undetected outlier sample.
Detailed Description
The image analysis system of the invention comprises the following links: image acquisition, foreground point marking, adjacent frame change detection, suspected lump form filtering, image feature extraction, outlier sample filtering and result visualization.
Example 1:
(a) Scanning a human body through medical imaging equipment to obtain a CT image sequence which is arranged in sequence, wherein two slices which are adjacent in space correspond to images of adjacent frames;
(b) Marking the area occupied by the tumor at the corresponding coordinate position of the corresponding image frame where the suspected lung tumor is located;
(c) Extracting image features for each suspected mass:
calculating the mean value of CT values of the suspected lump area as the characteristic of the area;
calculating the standard deviation of the CT value of the suspected lump area as the characteristic of the area;
(d) Analyzing the image characteristics of all suspected tumor areas of all patients based on an outlier sample detection (outlier detection) technology, and filtering out the suspected tumor areas in an image set corresponding to the feature space outlier samples;
(e) And marking the lump at the corresponding coordinate position of the corresponding image frame.
In the step (d), the detection method of the outlier sample comprises the following steps:
(d1) Pretreatment: assuming that there are L suspected masses, the feature corresponding to each mass is a vector, and the feature space has L samples { f } 1 ,f 2 ,…,f L Normalizing for each characteristic dimension such that the values of the L samples in the characteristic dimension obey a standard normal distribution;
(d2) Judging an outlier sample by adopting a K nearest neighbor method: sample f i Average distance O to its K nearest neighbors i As an abnormality metric index (i =1,2, …, L) { O 1 ,O 2 ,…,O L Sorting from large to small, and taking the corresponding first L' samples as outlier samples, where O L′ Is { O 1 ,O 2 ,…,O L Where an inflection point appears after the ordering.
Here, the inflection point is calculated by referring to [ Su Yang, zhishun Li, and Xinlong Wang, "Vessel radial noise recognition with fractional defects", IEE Electronics Letters, vol.36, no.10, pp.923-925,2000].
In step (b), the lung mass is determined (segmented and located) as follows:
(b1) For an image subsequence containing lungs, marking foreground points and background points, wherein points corresponding to the area where the lungs of each frame of image are located are used as the foreground points, and the rest points are used as the background points;
(b2) Calculating the change between adjacent frame images, and marking the mutation area as a suspected lump: order to
Indicating that the image of the t-th frame is located at the coordinate (x) i ,y i ) The point (b) of,
And
respectively representing whether the point belongs to the foreground or the background, observing K +1 continuous frame images (K is more than or equal to 1), and regarding a point set meeting the following formula as a suspicious point:
clustering the obtained suspicious points into connected regions by using an 8-neighbor region growing method, wherein each obtained connected region corresponds to one suspected lump;
(b3) Filtering the suspected mass:
if the number of points contained in a connected region consisting of foreground points is less than 0.0005 xMxN, where M and N represent the height and width of the image, excluding the region from the set of suspected tumor regions;
performing principal component analysis (principal component analysis) on a connected region formed by a foreground point, dividing the larger value of the two obtained characteristic values by the smaller value, and if the ratio is larger than a threshold value 3, excluding the region from the set of suspected lump regions;
if the ratio of the number of points contained in a connected region formed by a foreground point to the area of the minimum circumscribed rectangle of the region is less than 0.35, the region is excluded from the set of suspected lump regions.
The lung mass segmentation and localization method of step (b) comprises the following steps of calculating step (b 1):
(1) Assuming that the resolution of each slice image is M multiplied by N, and the CT value of a pixel (x, y) is I (x, y), marking the pixel points of which the CT value in the image satisfies {74 ≦ I (x, y) ≦ 774 x ∈ [1,M ] ^ y ∈ [1,N ] };
(2) Clustering the foreground points into connected regions by an 8-neighbor region growing method, wherein the definition of the connectivity is as follows: if some of 8 neighbor points of a foreground point are also foreground points, the neighbor points are communicated with the point and belong to the same communicated area;
(3) Filtering and screening the connected region by adopting a filtering method: if the number of points contained in a connected region formed by a foreground point is less than 0.01 multiplied by M multiplied by N, excluding the region from the set of connected regions;
(4) Sequentially scanning (in the direction from head to foot of the human body) the CT image of each slice until an image is found which contains and only contains 2 connected regions as the starting frame of the image subsequence of lungs, while marking the two connected regions as left and right lungs, respectively, according to position;
(5) Scanning downwards from the left lung and the right lung along the initial frame respectively to find a current lung region meeting the consistency test with the previous frame of image; the consistency checking method comprises the following steps:
let n be i Representing the number of foreground points contained in a lung region in an image of a current frame (i-th frame), nOld representing the number of foreground points in the lung region of a previous frame, and nMax representing the number of foreground points corresponding to the maximum lung region recorded until the current frame is scanned;
if n is i <1.5 × nOld, the current frame passes consistency check, and nOld and nMax are updated:
let nOld = n i ；
If n is i >nMax, let nMax = n i ；
Otherwise, recording the current frame number T, and turning to the step (7);
(6) Repeating the step (5) until n i <0.5 multiplied by nMax, and recording the current frame number T;
(7) Tracing back from the current frame T until n appears i ≥n i-1 I-1 is marked as the last frame of the sub-sequence of lung images.
In the step (2), the calculation steps of the 8-neighbor region growing method are as follows:
(1) randomly selecting one foreground point which is not scanned as a seed point of a current connected region;
(2) scanning 8 neighbor points of each foreground point contained in the current connected region and adding the foreground points into the current connected region;
(3) repeating the step (2) until no more foreground points communicated with the area can be found;
(4) if all foreground points are scanned, ending; otherwise, turning to the step (1);
an experiment was conducted according to the method of example 1, and CT images of 6 patients were analyzed, and 11 outlier samples were found from the 29 detected masses, and after excluding these 11 false alarms, 8 of the remaining 18 samples were true masses, corresponding to 5 patients. The distribution of 29 masses in the feature space after image feature extraction is shown in fig. 1, and it can be seen that the distribution of the detected 11 outlier samples is relatively random, and no cluster with high consistency is formed. The 8 correctly detected tumors are shown in fig. 2. An example of a false-detected mass corresponding to an outlier sample is shown in fig. 3. Some examples of false positives after outlier rejection are shown in figure 4.
Claims (5)
1. A cross-image set outlier sample finding method for filtering lung mass misdetection results is characterized in that outlier sample detection is performed on a region which is judged to be a suspected mass in an earlier stage in a feature space by traversing the whole image set, and the method comprises the following specific steps:
(a) Scanning a human body through medical imaging equipment to obtain a CT image sequence which is arranged in sequence, wherein two slices which are adjacent in space correspond to images of adjacent frames;
(b) Marking the area occupied by the tumor at the corresponding coordinate position of the corresponding image frame where the suspected lung tumor is located;
(c) Image features were extracted for each suspected mass:
(d) Analyzing the image characteristics of all suspected tumor areas of all patients by adopting an outlier sample detection technology, and filtering the suspected tumor areas in an image set corresponding to the feature space outlier sample;
in the step (d), one of the following 2 calculation methods is adopted:
the first calculation method comprises the following steps:
(1) Pretreatment: assuming that there are L suspected masses, the feature corresponding to each mass is a vector, and the feature space has L samples { f } 1 ,f 2 ,…,f L Normalizing for each characteristic dimension such that the values of the L samples in the characteristic dimension obey a standard normal distribution;
(2) Judging an outlier sample by adopting a K nearest neighbor method: sample f i Average distance O to its K nearest neighbors i As an anomaly metric index, i =1,2, …, L, will be { O 1 ,O 2 ,…,O L Sorting from large to small, and taking the corresponding first L' samples as outlier samples, where O L′ Is { O 1 ,O 2 ,…,O L Place where the inflection point appears after sorting;
the second calculation method comprises the following steps:
(1) Pretreatment: suppose there are L suspected masses, each mass has a corresponding feature as a vector, and the feature space has L samples { f } 1 ,f 2 ,…,f L Normalizing for each characteristic dimension such that the values of the L samples at the characteristic dimension obey a standard normal distribution;
(2) Establishing a connectivity graph between samples: calculating Euclidean distance between every two L samples:
{d ij |i＝1,2,…,L；j＝1,2,…,L},
to { d ij I =1,2, …, L; j =1,2, …, L } order, finding the value d at which the sequence inflection point occurs T And calculating:
in the above formula, i =1,2, …, L, j =1,2, …, L; if L samples are considered as L nodes constituting the graph model, r ij =1 denotes the communication between sample i and sample j, r ij =0 means that both are not connected; { r ij I =1,2, …, L; the connected relation among all samples recorded by j =1,2, …, L } forms a graph model, all connected subgraphs are found, and the samples contained in all subgraphs except the largest connected subgraph are outlier samples;
in step (b), the lung mass is determined by the following method:
marking foreground points and background points for an image subsequence containing lungs, wherein points corresponding to the area where the lungs of each frame of image are located are used as the foreground points, and the rest points are used as the background points;
(II) calculating the change between adjacent frame images, and marking the mutation area as a suspected lump: order to
Indicating that the image of the t-th frame is located at the coordinate (x) i ,y i ) The point (c) of (a) is,
and
respectively representing whether the point belongs to the foreground or the background, observing K +1 continuous frame images, wherein K is more than or equal to 1, and regarding a point set meeting the following formula as a suspicious point:
clustering the obtained suspicious points into connected areas by an 8-neighbor area growing method, wherein each obtained connected area corresponds to one suspected lump;
(III) filtering the suspected tumor:
if the number of points contained in a connected region consisting of foreground points is less than 0.0005 xMxN, where M and N represent the height and width of the image, excluding the region from the set of suspected tumor regions;
performing principal component analysis on a connected region formed by a foreground point, dividing a larger value of the two obtained characteristic values by a smaller value, and if the ratio is greater than a threshold value, excluding the region from the set of suspected lump regions;
if the ratio of the number of points contained in a connected region consisting of a foreground point to the area of the smallest bounding rectangle of the region is less than a certain threshold, the region is excluded from the set of suspected mass regions.
2. The method of claim 1, wherein the extracted image features in step (c) are:
calculating the mean value of CT values of the suspected lump area as the characteristic of the area;
alternatively, the standard deviation of the CT value is calculated for the suspected mass region as the feature of the region.
3. The method of claim 1, wherein the calculating step of step (one) is as follows:
(1) Assuming that the resolution of each slice image is M multiplied by N, and the CT value of a pixel (x, y) is I (x, y), marking the pixel points of which the CT value in the image satisfies {74 ≦ I (x, y) ≦ 774 x ∈ [1,M ] ^ y ∈ [1,N ] };
(2) Clustering the foreground points into connected regions by an 8-neighbor region growing method, wherein the definition of the connectivity is as follows: if some of 8 neighbor points of a foreground point are also foreground points, the neighbor points are communicated with the point and belong to the same communicated area;
(3) Filtering and screening the connected region by adopting a filtering method: if the number of points contained in a connected region formed by a foreground point is less than 0.01 multiplied by M multiplied by N, excluding the region from the set of connected regions;
(4) Sequentially scanning the CT image of each slice until an image is found that contains and only contains 2 connected regions as the starting frame of the lung image sub-sequence, while labeling the two connected regions as left and right lungs, respectively, according to position;
(5) Scanning downwards from the left lung and the right lung along the initial frame respectively to find a current lung region meeting the consistency test with the previous frame of image; the consistency checking method comprises the following steps:
let n be i Representing the number of foreground points contained in a lung region in an image of a current frame (i-th frame), nOld representing the number of foreground points in the lung region of a previous frame, and nMax representing the number of foreground points corresponding to the maximum lung region recorded until the current frame is scanned;
if n is i <1.5 × nOld, the current frame passes consistency check, and nOld and nMax are updated:
let nOld = n i ；
If n is i >nMax, then let nMax = n i ；
Otherwise, recording the current frame number T, and turning to (7);
(6) Repeating the step (5) until n i <0.5 multiplied by nMax, and recording the current frame number T;
(7) Tracing back from the current frame T until n appears i ≥n i-1 I-1 is marked as the last frame of the sub-sequence of lung images.
4. The method of claim 3, wherein the 8-neighbor region growing method in step (2) is calculated by:
(1) randomly selecting one of the foreground points which are not scanned as a seed point of the current connected region;
(2) scanning 8 neighbor points of each foreground point contained in the current connected region and adding the foreground points into the current connected region;
(3) repeating the step (2) until no more foreground points communicated with the area can be found;
(4) if all foreground points are scanned, ending; otherwise, turning to the step (1).
5. An outlier sample finding system across a set of images for filtering lung mass misdetection results based on the outlier sample finding method according to one of the claims 1-4, comprising 4 modules: the system comprises a human body image sequence acquisition and sorting module, a suspected lung lump region marking module, a suspected lump extraction image feature module and an outlier sample detection module; these 4 modules in turn correspond to the operational content of steps (a), (b), (c), (d) of the cross-image set outlier sample detection method for filtering lung mass misdetection results.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910436092.0A CN110349662B (en) | 2019-05-23 | 2019-05-23 | Cross-image set outlier sample discovery method and system for filtering lung mass misdetection results |
| PCT/CN2019/092043 WO2020232786A1 (en) | 2019-05-23 | 2019-06-20 | Image set-based outlier sample discovery method for filtering out false pulmonary mass detection results, and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910436092.0A CN110349662B (en) | 2019-05-23 | 2019-05-23 | Cross-image set outlier sample discovery method and system for filtering lung mass misdetection results |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110349662A CN110349662A (en) | 2019-10-18 |
| CN110349662B true CN110349662B (en) | 2023-01-13 |
Family
ID=68174294
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910436092.0A Active CN110349662B (en) | 2019-05-23 | 2019-05-23 | Cross-image set outlier sample discovery method and system for filtering lung mass misdetection results |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110349662B (en) |
| WO (1) | WO2020232786A1 (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112529918B (en) * | 2020-12-23 | 2024-02-27 | 沈阳东软智能医疗科技研究院有限公司 | Method, device and equipment for segmenting brain room area in brain CT image |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101103924A (en) * | 2007-07-13 | 2008-01-16 | 华中科技大学 | Galactophore cancer computer auxiliary diagnosis method based on galactophore X-ray radiography and system thereof |
| WO2012024258A1 (en) * | 2010-08-17 | 2012-02-23 | Bank Of America Corporation | Systems and methods for performing access entitlement reviews |
| CN105138641A (en) * | 2015-08-24 | 2015-12-09 | 河海大学 | Angle-based high dimensional data outlier detection method |
| CN105427325A (en) * | 2015-12-07 | 2016-03-23 | 苏州大学 | Automatic lung tumour segmentation method based on random forest and monotonically decreasing function |
| CN108416190A (en) * | 2018-02-11 | 2018-08-17 | 广州市碳码科技有限责任公司 | Tumour methods for screening, device, equipment and medium based on deep learning |
| CN108537773A (en) * | 2018-02-11 | 2018-09-14 | 中国科学院苏州生物医学工程技术研究所 | Intelligence auxiliary mirror method for distinguishing is carried out for cancer of pancreas and pancreas inflammatory disease |
| CN108596770A (en) * | 2017-12-29 | 2018-09-28 | 山大地纬软件股份有限公司 | Medicare fraud detection device and method based on outlier analysis |
| CN108703741A (en) * | 2018-03-30 | 2018-10-26 | 西安交通大学 | A kind of shifty NIRS Interference Detections and minimizing technology |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8238637B2 (en) * | 2006-10-25 | 2012-08-07 | Siemens Computer Aided Diagnosis Ltd. | Computer-aided diagnosis of malignancies of suspect regions and false positives in images |
| US9661299B2 (en) * | 2011-06-30 | 2017-05-23 | Thomson Licensing | Outlier detection for colour mapping |
| CN103400385B (en) * | 2013-07-22 | 2016-05-25 | 西安电子科技大学 | The doubtful lymph node extracting method of stomach CT image that sparse dynamic integrity is selected |
-
2019
- 2019-05-23 CN CN201910436092.0A patent/CN110349662B/en active Active
- 2019-06-20 WO PCT/CN2019/092043 patent/WO2020232786A1/en active Application Filing
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101103924A (en) * | 2007-07-13 | 2008-01-16 | 华中科技大学 | Galactophore cancer computer auxiliary diagnosis method based on galactophore X-ray radiography and system thereof |
| WO2012024258A1 (en) * | 2010-08-17 | 2012-02-23 | Bank Of America Corporation | Systems and methods for performing access entitlement reviews |
| CN105138641A (en) * | 2015-08-24 | 2015-12-09 | 河海大学 | Angle-based high dimensional data outlier detection method |
| CN105427325A (en) * | 2015-12-07 | 2016-03-23 | 苏州大学 | Automatic lung tumour segmentation method based on random forest and monotonically decreasing function |
| CN108596770A (en) * | 2017-12-29 | 2018-09-28 | 山大地纬软件股份有限公司 | Medicare fraud detection device and method based on outlier analysis |
| CN108416190A (en) * | 2018-02-11 | 2018-08-17 | 广州市碳码科技有限责任公司 | Tumour methods for screening, device, equipment and medium based on deep learning |
| CN108537773A (en) * | 2018-02-11 | 2018-09-14 | 中国科学院苏州生物医学工程技术研究所 | Intelligence auxiliary mirror method for distinguishing is carried out for cancer of pancreas and pancreas inflammatory disease |
| CN108703741A (en) * | 2018-03-30 | 2018-10-26 | 西安交通大学 | A kind of shifty NIRS Interference Detections and minimizing technology |
Non-Patent Citations (1)
| Title |
|---|
| 基于高斯核函数的局部离群点检测算法;关绍云等;《哈尔滨商业大学学报( 自然科学版)》;20190430;第185-203页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110349662A (en) | 2019-10-18 |
| WO2020232786A1 (en) | 2020-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zebari et al. | Improved threshold based and trainable fully automated segmentation for breast cancer boundary and pectoral muscle in mammogram images | |
| CN107067402B (en) | Medical image processing apparatus and breast image processing method thereof | |
| Zhang et al. | Automated semantic segmentation of red blood cells for sickle cell disease | |
| US20110172514A1 (en) | Method for increasing the robustness of computer-aided diagnosis to image processing uncertainties | |
| KR20160063128A (en) | Apparatus and Method for Computer Aided Diagnosis | |
| CN111986189B (en) | Multi-category pneumonia screening deep learning device based on CT images | |
| Xu et al. | Computerized classification of prostate cancer gleason scores from whole slide images | |
| Naresh et al. | Early detection of lung cancer using neural network techniques | |
| US7555152B2 (en) | System and method for detecting ground glass nodules in medical images | |
| CN113256634B (en) | Cervical carcinoma TCT slice vagina arranging method and system based on deep learning | |
| Abdullah et al. | Multi-sectional views textural based SVM for MS lesion segmentation in multi-channels MRIs | |
| CN110349662B (en) | Cross-image set outlier sample discovery method and system for filtering lung mass misdetection results | |
| Nobile et al. | Unsupervised neural networks as a support tool for pathology diagnosis in MALDI-MSI experiments: A case study on thyroid biopsies | |
| CN104835155A (en) | Fractal-based early-stage breast cancer calcification point computer auxiliary detection method | |
| Basheer et al. | Classification of breast masses in digital mammograms using support vector machines | |
| Vani et al. | Automatic tumor classification of brain mri images | |
| Gomathi et al. | Automated CAD for lung nodule detection using CT scans | |
| Herwanto et al. | Association technique based on classification for classifying microcalcification and mass in mammogram | |
| Yao et al. | Computer aided lytic bone metastasis detection using regular CT images | |
| Ganeshkumar et al. | Unsupervised deep learning-based disease diagnosis using medical images | |
| CN111311589A (en) | Pulmonary nodule detection and property judgment system and method | |
| Abdullah et al. | Lung Lesion Identification Using Geometrical Feature and Optical Flow Method from Computed Tomography Scan Images | |
| Azemin et al. | Unsupervised Chest X-ray Opacity Classification using Minimal Deep Features | |
| D’cruz et al. | Brain abnormality detection in MRI images based on estimation of statistical texture measures | |
| Pradipkumar et al. | Segmentation of large scale medical images using HPC: Classification of methods and challenges |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |