Method for obtaining pathological diagnosis according to pulmonary nodule CT image by applying artificial intelligence technology

Abstract

The invention relates to a method for obtaining pathological diagnosis according to a pulmonary nodule CT image by applying an artificial intelligence technology. An algorithm model is constructed by applying an artificial intelligence technology to form a pulmonary nodule CT image-pathological diagnosis system, so that the pulmonary nodule CT image-pathological diagnosis system has the functions of image acquisition, image preprocessing, feature extraction, image recognition and the like, and CT image features of pulmonary nodules and internal features of abnormal cells in pathological tissue dyeing pictures of the pulmonary nodules can be recognized, extracted and memorized. CT image data of pulmonary nodules directly correspond to pathological picture data, and forward and reverse interpretation can be carried out on the data. A CT image database and a pathological picture database of pulmonary nodules are established, deep learning is carried out with sufficient data, and the functions of the pulmonary nodules are improved. The system is verified and clinically tested to ensure the accuracy of the system. After CT image data of pulmonary nodules are input into the system, pathological results can be displayed. The method has the advantages that the pathological result of the pulmonary nodule can be judged when the pulmonary nodule is seen on the chest CT, and the purpose of accurate treatment can be achieved.

Description

Technical Field

[0001] The present invention belongs to the field of medical technology and relates to the application of artificial intelligence in the fields of medical imaging and pathological diagnosis. In essence, it is to invent a method for obtaining the pathological diagnosis result based on the CT image of pulmonary nodules. Background Art

[0002] Pulmonary nodules are nodular lesions with a diameter less than 30 mm in the lung, and are mostly detected during chest CT examinations. The morphology of pulmonary nodules can be manifested as pure ground-glass nodules, mixed ground-glass nodules, and solid nodules on CT images. At the same time, these morphological manifestations also represent different pathological progression processes of pulmonary nodules, including adenomatous atypical hyperplasia, carcinoma in situ, microinvasive carcinoma, invasive carcinoma, and benign lesions, etc. Since different intervention methods need to be taken for pulmonary nodules at different pathological stages, it is crucial to master which pathological stage of lung cancer the CT image of the pulmonary nodule is in or whether it is a benign nodule.

[0003] The imaging diagnosis of pulmonary nodules is currently in a relatively limited state. Usually, doctors study the CT images of pulmonary nodules and guess the possible pathological results based on experience, and then decide whether to perform surgery and what surgical method to take if surgery is needed. However, this method cannot meet the clinical doctors' control of the intervention timing of pulmonary nodules. In some cases, the surgical method is determined based on frozen pathology during the operation, which is cumbersome and time-consuming. In some cases, a few days after the operation, when the formal pathological report comes out, it is found that there are differences in the pathological classification or pathological subtype level from the frozen pathological report during the operation. At this time, the patient may need a second operation or additional adjuvant treatment. This affects the treatment effect, increases the patient's pain, and wastes medical expenses. In addition, in some cases, surgery may be performed due to premature selection of the surgical timing or unnecessary intervention, resulting in over-treatment. Therefore, it is urgently needed for us to create a new solution for the diagnosis of pulmonary nodules based on imaging and pathology.

[0004] Artificial intelligence technology has been very mature. Relying on computer vision technology to process, analyze, and understand images, image recognition technology has been formed. By applying this technology to build models and create algorithms, through steps such as image acquisition, image preprocessing, feature extraction, and image recognition, a medical imaging artificial intelligence analysis system has been established and applied to the imaging diagnosis in the radiology department. Traditionally, for the tissue specimens removed after surgery, the pathologist observes the stained sections through a microscope with the naked eye and makes a pathological diagnosis. With the progress of computer vision technology, computational pathology has been created. Research shows that by analyzing digitized pathological tissue staining pictures and performing machine learning under the condition of large-scale data, the artificial intelligence pathological diagnosis system has extremely high accuracy and reliability in the diagnosis of cancer tissue specimens.

[0005] However, at present, the medical artificial intelligence imaging diagnosis system and the pathological diagnosis after surgery exist independently of each other, and it is not yet possible to know the pathological results as soon as a lesion is seen on the image, which is exactly the future development direction. Therefore, the present invention is to connect the artificial intelligence imaging diagnosis system for diagnosing pulmonary nodules with the pathological diagnosis after pulmonary nodule surgery and the related artificial intelligence pathological diagnosis system, so as to match the corresponding pathological results when seeing the CT image of the pulmonary nodule, and then achieve the purpose of precise treatment. Summary of the Invention

[0006] To solve the existing problems, the present invention provides a solution, that is, to apply artificial intelligence technology to create a method for obtaining a pathological diagnosis based on the CT images of pulmonary nodules, that is, to know the pathological results as soon as a pulmonary nodule is seen. The summary of the invention includes: 1. Design a software diagnosis system for CT images - pathology of pulmonary nodules, solve the comprehensive algorithm problem, and be able to perform deep learning; 2. This diagnosis system can identify the CT image features of pulmonary nodules and the abnormal cell features in the pathological tissue staining pictures thereof, and establish an association between the two. 3. This diagnosis system has the functions of forward and reverse interpretation and memory function for the CT image data and pathological tissue picture data of pulmonary nodules; 4. Establish a CT image database of pulmonary nodules and its pathological tissue staining picture database to have sufficient data for machine learning; 5. This diagnosis system has the ability to obtain pathological results as soon as a pulmonary nodule is seen on a chest CT; 6. The pathological results diagnosed by this system will be displayed in the form of graphics or text.

[0007] Compared with the prior art, the present invention has the following advantages: This diagnosis system is a new diagnosis technology invented by applying artificial intelligence technology based on deep learning of a large amount of CT image data and pathological tissue picture data of pulmonary nodules; it can achieve the purpose of directly obtaining pathological diagnosis results based on the CT images of pulmonary nodules; since the pathological process stage of pulmonary nodules is clearly mastered, appropriate intervention methods can be adopted to perform precise treatment, which can not only prevent missed diagnosis but also prevent over-treatment.

[0008] The technical solution is as follows: 1. Establish an artificial intelligence CT image-pathological diagnosis system for pulmonary nodules. Apply machine learning algorithms, natural language processing algorithms, and computer vision algorithms of artificial intelligence to construct a new algorithm model, associate the CT image features of pulmonary nodules with the abnormal cell features in their pathological pictures, and form a CT image-pathological diagnosis system for pulmonary nodules; 2. This diagnosis system has functions such as image acquisition, image preprocessing, feature extraction, and image recognition. It can identify, extract, and memorize the CT image features of pulmonary nodules and the internal features of abnormal cells in their pathological tissue staining pictures, and can also interpret the CT image data and pathological picture data of pulmonary nodules in a forward and reverse manner; 3. Different morphological features of pulmonary nodules on chest CT images are pathologically associated with adenomatous atypical hyperplasia, in-situ carcinoma, micro-invasive carcinoma, invasive carcinoma, and benign lesions, etc. After the CT image data of the pulmonary nodules of each patient is input, the pathological results will be displayed simultaneously; 4. Establish a chest CT image database of 100,000 patients with pulmonary nodules who have undergone surgery and a pathological tissue picture database of 100,000 corresponding patients. When each CT image data of a pulmonary nodule is input, the pathological picture data of this patient is input at the same time. Associate the CT image features of the pulmonary nodules with the internal features of abnormal cells in the pathological tissue staining pictures, and perform deep learning to improve its function; 5. The verification database is used to detect the performance of this diagnosis system: 10,000 cases of CT images of pulmonary nodules of patients and the clinical pathological reports of the corresponding patients who have undergone surgery are required. That is, input the CT image data of a pulmonary nodule of a certain patient into this diagnosis system, and verify whether the displayed pathological report is consistent with the clinical pathological results to ensure its accuracy on unseen data; 6. The test database also requires 10,000 patients. Predict the pathological results based on the CT images of the pulmonary nodules of the patients before surgery and compare them with the clinical pathological reports after surgery to evaluate the accuracy of this diagnosis system; 7. Present the diagnostic results made by the artificial intelligence CT image-pathological diagnosis system for pulmonary nodules in graphical or text form.

[0009] Description of the embodiment: In a hospital or physical examination center, when a pulmonary nodule is found on a patient's chest CT, the CT image data of this pulmonary nodule is immediately input into the artificial intelligence CT image-pathological diagnosis system for pulmonary nodules. Based on the CT image features of the pulmonary nodule, this diagnosis system will find the corresponding pathological features to make a pathological diagnosis and display the results, which serves as evidence for whether to intervene in the pulmonary nodule clinically.

Claims

1. Use artificial intelligence technology to create a method for pathological diagnosis based on CT images of lung nodules, which is characterized by Supported by artificial intelligence technology, the system can make a pathological diagnosis when lung nodules are seen on chest CT.

2. The diagnostic system according to claim 1 is a pulmonary nodule CT image-pathology diagnostic system formed by constructing an algorithm model using artificial intelligence machine learning algorithms, natural language processing algorithms, and computer vision algorithms. The diagnostic system has functions such as image acquisition, image preprocessing, feature extraction, and image recognition.

3. The diagnostic system according to claims 1 and 2 can identify, extract and memorize the CT image features of lung nodules and the internal features of abnormal cells in their pathological tissue staining images, and can also perform forward and reverse interpretation of the CT image data and pathological image data of lung nodules.

4. The diagnostic system according to any of the above claims, wherein the diagnostic result is displayed in graphical or textual form.