JP6503535B1 - A diagnostic method of displaying medical images on an image by symptom level at the judgment of an AI. - Google Patents

Abstract

【Task】
The diagnosis of the current medical image scan is judged by the doctor visually, and there is doubt about the current diagnosis method that the patient or the examiner listens to the doctor's response, and it is required to judge by medical science means. It was an issue to diagnose from the fact that it utilized AI.
[Solution means]
The diagnostic method according to the present invention combines the machine learning of medical image scanning with AI and the diagnostic method of judging the degree of symptoms by numerical value and displaying it on an image, accurate diagnosis based on modern medical science is a request of the times and The conventional problem is solved by using a diagnostic method.

Description

  In the present invention, conventionally, in medical examination for medical scan images, it is common to visually judge the degree of symptoms by visual inspection, and a doctor who examines an image that appears delicately and finely while busy to judge the degree of symptoms The method of diagnosing with one's own knowledge and experience is a technical field that is medically problematic and ameliorating.

  In today's medical engineering, taking advantage of AI and imaging of a region instructed by a doctor, the image scan can be made for the site symptoms other than the purpose and the degree of symptoms can be determined by the AI judgment. The present visual diagnosis method can be improved by applying this technology to medicine.

  The problem can be solved by improving misdiagnosis and oversight by the diagnostic method of judging medical images of the present invention by AI and displaying the degree of symptoms on the images.

  The present invention relates to a diagnostic method of judging a medical image devised in view of such a situation with AI and displaying the degree of symptoms on the image.

In the medical image according to claim 1, in addition to the conventional visual and diagnostic interpretation, in the diagnosis of artificial intelligence (referred to as "AI") utilization, AI based on the past case images for each medical imaging form, We made machine learning of artificially created teacher data artificially created teacher data that entered numerical values indicating the degree of progress in the circled shape by manipulating the symptom site of the photographed image in a circular shape by a specialist doctor (hereinafter referred to as "teacher") prevented by software, the captured image of the image region that the physician instructs when animal screening and medical examination without the human, site conditions other than indicated can also be found in the judgment of the AI, yet overlooked symptoms determination of the image, or a misdiagnosis a medical image, characterized in AI diagnostics that diagnostic method for displaying the symptoms degree each image at the discretion of the AI.

The diagnostic method of judging medical images according to the present invention by AI and displaying the degree of symptoms on an image can prevent misdiagnosis or overlook of subtle image symptom judgment, and also use AI that patients and relatives can be trusted in diagnosis results According to the present invention, quick and accurate medical examination and medical examination can be performed, and effects such as reduction of labor of a doctor can be expected at a medical site by using an apparatus incorporating software according to the present invention.

In the DICOM standard, medical images taken with digital radiography, DR, CT, MRI, echo, etc. imaging devices (this is called "modality" in the medical field) are first of all in file format according to the DICOM standard. It is reserved for storage (storage area, usually hard disk) in each modality. Images are usually transferred (copied) to an image database called PACS and can be written about PACS.
If the information image CD for DICOM image viewer is a PC with Windows (Windows), insert the CD into the drive, the viewer program is automatically started, and if you operate it according to the explanation displayed on the screen, you can see the image It is supposed to be able to
The DICOM standard is an international standard for medical imaging and communication established by the Radiological Society of the United States. It is a storage format for all digital medical imaging data handled in the medical field as well as radiography, and also , A communication protocol for devices that handle those images. In other words, it is a standard for medical images and communication, and it is used to display medical images on various medical devices, burn image data and viewers on CDs, and provide diagnostic information to other clinics.

The present invention is a medical technology that diagnoses whether there is an abnormality or not by imaging an internal disease or condition (a tumor, an infarction, an aneurysm, etc.) which can not be understood only by examining from outside the body with regard to diagnostic imaging and examination. There are PET-CT examinations, MRI examinations, and CT examinations as devices for obtaining images in the body, and diseases such as cancer can be detected at an early stage.
A radiologist is a doctor who has trained for more than 5 years at a facility designated by the Japanese Society for Medical Radiology and has passed the examination. It includes various examinations such as X-ray, CT, MRI, and ultrasound. From the diagnostic imaging methods, we select the test method that is appropriate for the symptoms and suspected disease, and the expert's eye interprets the test image that has been taken.

In the present invention, medical image scanning means to scan, examine, examine, look over, examine in detail, etc., for example, nuclear magnetic resonance imaging (MRI image), positron tomography (PET image) , X-ray projection, linear tomography, PoIy tomography, angulation tomography, orthopantomography (OPT image), computed tomography (CT image), etc., among medical examination contents, internal medicine, surgery, orthopedics, cranial nerve Scanning an image due to a disease at surgery, cranial nerve department, etc. is called medical image processing.

Medical imaging refers to techniques and processes that generate images of the human body for disease diagnosis and testing and medicine. It is part of biomedical photography that is not limited to humans, and is closely related to radiology, endoscopy, thermography, medical photography, microscopy, and so on. Measurement methods and recording methods (electroencephalograms and magnetoencephalograms) that were not originally designed to generate images can be viewed as a form of medical imaging because they generate data that can be represented like a kind of map. For medical images handled in diagnostic imaging, a doctor who medically interprets the image in the above paragraph 0007 is referred to as a radiodiagnostic doctor or an imaging diagnostician, which is one of the specialties of the doctor.
It is a field of medical imaging to apply image processing as needed to the image taken, and in medical facilities, radiologists (meaning for leading-edge imaging diagnosis) or medical radiologists can do especially. Although there are many, it is also possible to view the imaging form of the various image processing described above as imaging.

Medical imaging is often regarded as a technique for generating an image that visualizes the inside of the human body for the technical meaning of "medical imaging form". For example, in the case of ultrasound examination, it is possible to generate ultrasound in tissue. You can know the internal structure from the echo of. In the case of X-rays, images are obtained utilizing the fact that the absorptivity of X-rays differs in bones, fats, etc. That is, this is a technique in the form of CT, MRI, or an ultrasound image.
Also, as described above, recently, software for converting CT, MRI, and ultrasound images into three-dimensional images has appeared, and CT and MRI are originally two-dimensional images, as described above. Was projected on the film. A plurality of shots are taken to generate a three-dimensional image, and integration and three-dimensional modeling are performed using their computers. Three-dimensional ultrasound images are also generated in a similar manner. Because such important structures can be visualized in detail, three-dimensional visualization methods have become important information sources due to various diagnostic and surgical treatment forms.

Software is a term that is used in the computer field as opposed to hardware (physical machine), and refers to a computer program that performs some kind of processing, as well as related documents. Software generally includes application software (application software and applications) for specific tasks and tasks such as word processing software, and an operating system (OS that provides hardware management and basic processing to application software and users. Etc.) are classified as system software.
Software creation is "programming (computer)" and "software engineering" is an act in the field of systematically and quantitatively examining its application with respect to software development, operation and maintenance.

Recently, software has been introduced that converts CT, MRI, and ultrasound images into three-dimensional images. CT and MRI originally used to display a two-dimensional image on a film. In order to generate a three-dimensional image, multiple shots are taken and integrated using these computers for three-dimensional modeling. Three-dimensional ultrasound images are also generated in a similar manner.

Artificial intelligence (AI) refers to a field of computer science that studies "intelligence" using the concept of "calculation" and the tool of "computer". "Technology that allows computers to perform intellectual actions such as language understanding and reasoning and problem solving on behalf of human beings" or "research field on design and realization of intelligent information processing system by computer" Ru.
Experts say: In other words, if you are simply misunderstood and rephrased in plain English, “what intellectual actions (recognition, reasoning, language management, creation, etc.) that could only be done until now, such as what kind of procedure (algorithm) and what kind of data If you prepare information and knowledge, you can do it mechanically.

What makes AI learn is the most typical answer, for example, to learn how to distinguish "dogs" or "cats" (or different creatures) when looking at a photo.
Even if you understand at a glance human beings do not know AI that has not learned anything, you can not derive an appropriate answer without teaching.
There are two major types of techniques for learning AI. What kinds of technologies are there such as “machine learning” and “deep learning” that can be used to improve system efficiency and speed up data analysis?
Let's look at "machine learning" first. The technology has the feature that the developer (teacher) does not program all operations in advance, but AI analyzes the data itself and finds out the rules and rules. In other words, an AI that allows you to perform specific tasks by "training". For example, in the case of image recognition, AI is caused to read a large number of images of apples tagged with “red apple” and “blue apple” one by one. At that time, if you give instructions to AI by paying attention to the color, AI will focus on the color of the apple even when the image of the apple that has not been analyzed comes out. And they learn to distinguish between "red apple" and "blue apple".

Deep learning is a further development of machine learning. The major difference from conventional machine learning is that the framework used to analyze information and data is different. This is a "neural network" created by imitating human nerves, making computer analysis and learning of data powerful.
In order to know in more detail, we will analyze the image earlier and look at the mechanism to identify "blue apple" or "red apple". While machine learning had to be specified to focus on "color", AI learns by itself "dice spot" to distinguish in deep learning, and improves its performance . In other words, by watching a lot of data, Deep Learning learns on its own how to focus on it, and it automatically gets smarter without waiting for instructions from humans.
Although “machine learning” and “deep learning” are AI mechanisms, it can be said that the difference is that automation of functional enhancement is being promoted. In particular, it can be said that the system is evolving in that it automatically finds out "a place to be marked up (feature amount)" when distinguishing the object of analysis.

The workflow step (flow chart) of machine learning in the present application is created and operated in the following material (1), creation (2) to (5), and delivery operation (6).
(1) Preparation of data ・ ・ ・ Examination of what kind of data to use → Collection of data ・ What is data? Radiation image and endoscopic image taken by diagnosis of the patient in the past in the field of medical institution such as university hospital Image data such as ultrasound image, skin disease image, fundus examination image, CT, MRI, etc. Based on this, the purpose of the database is explained to the doctor in charge of image data for each specialized field, and the specialist in the field gets the cooperation There is a sufficient amount of data processing by entering a numerical value indicating the degree of progress in the area by circled by mouse operation for each symptom location by gray color of medical display screen (monochrome or color image) Collect image data that artificially created teacher data.

(2) Selection of method ... selection of "method / learning method / algorithm" of machine learning-Machine learning is one of the methods for creating artificial intelligence (AI). A normal machine works according to a program written by humans in detail. However, in machine learning, human beings do not create programs, but machines themselves learn from huge data and create models like judgment criteria. And I will give the answer myself. Machine learning is learning for that. Also, an algorithm is a procedure that solves a specific problem clearly defined as a combination of simple calculations and operations, and includes mathematical solutions and calculation procedures etc. However, IT (computer in the sense of information technology) In the field of “Technology related to data communication”, it often refers to a set of processing procedures formulated to be able to be given and executed in the form of a program in a computer.

(3) Preprocessing · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · data sorting Say.
・ Data expansion means movement, rotation, enlargement, reduction, distortion, etc. of the image of training data
A technique that increases the number of data by many times by performing operations such as noise addition.
・ Dividing data for learning and evaluation needs to be separated from training data in supervised learning.
Divide into 1, training data, 2, accuracy verification data, 3, test data.

(4) Model training · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · Trainers (trainers) handle the following three types of data when training (training / training) models.
Input data (training data) is a set of individual records (instances = examples) that contain features that are important for machine learning problems. This data is used for training, and the model is configured to make accurate predictions for new instances of similar data.
• Model parameters are variables used to adapt the selected machine learning method to the data. For example, the Deep Neural Network (DNN) consists of a large number of processing nodes (neurons = structural and functional units), and operations are defined for each node. As data moves through the network, operations on each node are performed on the data. When training a DNN, the weights defined for each node determine the magnitude of that node's impact on the final forecast. This weight is an example of a model parameter. It is a distinction between models with parameters and other models of the same type that operate on similar data.
• Hyperparameters are variables related to the training process itself, if the model parameters are variables and will be adjusted by training with existing data. For example, when setting up a deep neural network, determine the number of "hidden" layers of nodes to use between the input and output layers and the number of nodes to use for each layer. These variables are not directly related to training data. These are configuration variables. Another difference is that parameters are changed during the training job, but hyperparameters generally do not change during job execution.
Model parameters are optimized (referred to as "tuning") by the training process. Perform model operations on the data, compare the resulting predictions to the actual values of each data instance, evaluate the accuracy, and adjust until the best value is found. The adjustment of the hyperparameters is similar, running the entire training job, examining and adjusting the overall accuracy. In either case, make changes to the model configuration and find the best combination to handle the problem.
・ What is learning ... ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ The hyper-parameters of how to learn the data should be set by human. You usually do that and create a "learned model". This is the phase to learn AI. And using “learned models” to predict unknown events is a phase using AI, that is, a segment of the changing process.

(5) Evaluation of model ... inference → repeat (2) to (5) → select optimal model · Inference means to input data to a learned model and receive the result from that model as output . In other words, if it is supervised learning, use the accuracy verification data to check the accuracy of the model.
Evaluate the performance (generalization performance) of the learned model for unknown test data, and there is a possibility of bias problems due to data familiarity with accuracy verification data, so using fresh test data again, as much as possible in the operating environment Final check in close context. Repeat (2) to (5) to select an optimal model.
(6) Deploy learned models selected for delivery / operation selection to medical institutions etc. in the production environment, export the operation / selection selected learned models in a form that can be executed in the operation environment, and complete integration into AI services and applications .

As described above, it is necessary to prepare a sufficient amount of “teacher data” at the same time as speeding up the computer to put high-performance deep learning into practical use for using AI. Data is data that a computer learns, and its outcome determines the accuracy of deep learning.
In this case, teacher data to AI (machine learning) the image data of the present application refers to the shades of gray image of a medical display image scan (monochrome or color image) accumulated in the past diagnosis by a specialist engaged in each image field of the medical field. It is processing of the data which encloses in circle shape by mouse operation for each symptom part by color, and inputs the numerical value which shows the progress degree in the part, and specialists artificially made the correct answer, for example, the image is an initial symptom Humans (experts) will identify and define the correct answer as to whether they are late symptoms or not.
In this way, the teacher data is created by distributing a management viewer from a partner medical institution or a medical institution, or using a USB memory or data transmission etc. to provide a medical display image scan provided by a company specialist ( The teacher can process the data in the above-mentioned procedure and create the correct answer with the human tactics, but it can be very laborious.

That is, the specialist described in (1) examines the gray color condition of the monochrome or color image of the medical display image scan to determine the symptom location and the degree of the symptom, and the specialist specialist as a teacher clicks the PC in the image operation. If it is enclosed in a shape, and the degree of progress to the site, for example, the symptoms displayed in the image scan are examined and judged as the initial symptoms in two stages of "1-2" and judged as the intermediate symptoms Indicated in 2 stages of "3 to 4", when it is judged to be a late symptom, enter the numerical value shown in 2 stages of "5 to 6", and artificially process teacher data processed into data with finer stages of symptoms The image data created in is provided from the relevant medical institution, and the database collected is subjected to machine learning repeatedly in the workflow from (2) to (5), and an optimal model is selected. Therefore, since the above-mentioned learning system enables efficient image processing, it is an AI image of “calculation expression is unnecessary” in teacher data.
The potential benefits of AI are that the support of a specialist can reduce the effort and improve the efficiency of work, and also provide appropriate treatment based on quick and accurate diagnosis.

As application fields, a system for playing an intelligent game, an image recognition system (computer vision) for identifying an object or person in an image or video, a speech recognition system for listening to human speech and understanding of contents, assembling words It is known that various natural language processing such as voice synthesis systems, advanced autonomous control systems (automated driving etc.) automatic summarization and question answering systems for machines such as robots and automobiles, and advanced natural machine translation etc. This is used to determine the symptom determination / progress of medical image scanning. That is to collect the past symptom-by-disease symptom cases from the medical field and have the AI learn the symptoms for each medical condition by medical imaging form based on the example, and spread the diagnostic method by the AI.

  Although the diagnosis of medical image scan is judged by visual observation, however, he relies on the personal knowledge and experience of the doctor and examines the image of the scan for judgment, so the patient himself is listening to the doctor's way if it is the correct judgment. Is the present examination method. Therefore, because it is necessary to make medical judgment possible, it is in the utilization of artificial intelligence AI, and for example, a site instructed by a doctor by causing AI to machine learn symptoms of past medical conditions for each medical imaging form based on an example. In the image scan, it is also possible to newly detect a site symptom other than the instruction in the image scan by the judgment of the AI, and enclose the symptom site of the image in a circle and numerically indicate the symptom degree (for example Express the degree of being done in 6 levels of “1 to 6”), understand the numerical value, and notify the diagnosis by the method of making a voice by “voice system” to judge the symptom of the image regardless of the progress state of the symptom Misdiagnosis or missing can also be prevented, and when the image scan is examined and explained to the patient, the judgment value in the AI is used as the AI diagnosis method in which the patient and relatives are satisfied with the diagnosis result.

Image sites instructed by the doctor to AI, ie target symptoms of image scanning, include brain cancer, tongue cancer, laryngeal cancer, thyroid cancer, esophageal cancer, gastric cancer, colon cancer, gallbladder cancer, hepatocellular carcinoma, cholangiocarcinoma, pancreatic cancer, lung cancer , Breast cancer, ovarian cancer, cervical cancer, endometrial cancer, renal cell cancer, renal pelvic cancer, prostate cancer, bladder cancer, skin cancer, bone soft tissue tumor, malignant lymphoma, lip cancer, oral cancer, nose This includes pharyngeal cancer and childhood cancer.
Other types of stenosis include spinal stenosis, valvular heart disease, mitral stenosis, aortic stenosis, caries and periodontal disease, accident fractures, and "non-human" cats and Target images in animals including dogs,
These tests include "medical examination" and "examination" of "human-free" animals, and 1) medical examination means medical examination and 2) medical examination is an early stage of a specific disease. The aim is to discover and treat early.

  In order to judge the degree of symptoms of the site (stages "1, 2, 3, 4, 5, 6") by image scanning, specialized advanced knowledge and experience are also required, and it is particularly important to "Cancer" is the leading cause of death worldwide. World Health Organization (WHO) study is covering mortality, incidence, years of disability, years of loss, years of life with disability, disability adjusted years of life for 28 types of “cancer” in 188 countries from 1990 to 2013 To verify. Of the 14.9 million people who developed cancer in 2013, 8.2 million died. The number of years of life lost to health due to cancer, "the year of disability-adjusted life," is estimated to rise to 196.3 million years. The most common cancer in the world is estimated to be "prostate cancer" in men and "breast cancer" in women, and surpasses other "cancers." The most common deaths due to cancer in both sexes are 1.6 million deaths in the trachea, bronchi and lung cancer. Morbidity and mortality indicate that men and women, stomach cancer, liver cancer, esophagus cancer, cervical cancer, lip cancer, oral cancer, nasopharyngeal cancer are increasing in developing countries.

  From now on, “Health and disease free society to protect yourself” by medicine, from the engineering aspect “visualize the effect” by diagnosis of AI, show the tendency of future disease with numerical data by big data such as medical checkup and examination result, Condition of coexistence of illness and health, and even if it is somewhat ill, the unhealthy condition of working life actually takes quite a long time in life, but if you are not ill, how do you live while receiving the support of a medical and medical specialist In order to review the lifestyle by acting consciously, and to spread the health concept in a new longevity society without disease, it evolves into early predictive medicine that makes full use of technologies such as AI and VR (virtual reality and virtual reality). And have a long-term life expectancy that greatly changes with the idea of life-long service, actively engaging in "social activities and local communities", and having their own "value standards" to do their own things even in the years Social contribution Can.

The diagnostic method according to the present invention combines the machine-learning AI with the medical scan image and judges the symptom degree by a numerical value and displays it on the image by developing software that allows the AI to learn symptoms based on actual cases for each medical imaging form. It is possible for many people at the institution to extend the life expectancy by early detection of cancer etc. through medical examinations and medical examinations.

Claims (1)

In medical images, in addition to conventional visual and diagnostic radiology diagnosis, in the diagnosis of artificial intelligence (referred to as “AI”) utilization, a specialist doctor (hereinafter referred to as “teacher” based on past case images for each medical imaging form in AI. Does not include humans by software that artificially creates teacher data that artificially creates teacher data in which numerical values indicating progress are input to the circled shape by manipulating the symptom site of the captured image in a circular shape. in the captured image of the image portion the physician instructs when animal screening and medical examination, site conditions other than indicated can also be found in the judgment of the AI, yet overlooked symptoms determination of the image, or a misdiagnosis, characterized in AI diagnosis can be prevented diagnostic method for displaying the symptoms degree each image a medical image at the discretion of AI.