CN-122025072-A - Deep learning-based ovarian cancer diagnosis method and system

Abstract

The invention relates to the technical field of ovarian cancer diagnosis and discloses an ovarian cancer diagnosis method and system based on deep learning, wherein the method comprises the steps of establishing a diagnosis model according to historical diagnosis data of an ovarian cancer patient; the invention can comprehensively consider individual differences of different patients and various historical diagnosis strategies in the whole diagnosis period of ovarian cancer, thereby establishing a more accurate and comprehensive diagnosis model, outputting the comprehensive diagnosis result and carrying out dynamic evaluation, and updating the diagnosis result and the diagnosis model according to the evaluation result, so as to continuously improve the accuracy and reliability of diagnosis and effectively improve the clinical treatment effect of ovarian cancer.

Inventors

• LIU YINGLEI
• QIN JIARUI
• CAO YANG
• ZHANG WENJING

Assignees

• 南通市第一人民医院

Dates

Publication Date

20260512

Application Date

20251217

Claims (10)

1. 1. A deep learning-based ovarian cancer diagnostic method, comprising: establishing a diagnosis model according to historical diagnosis data of the ovarian cancer patient; Generating a diagnosis strategy of a patient to be diagnosed, and acquiring an image data packet of the patient to be diagnosed according to the diagnosis strategy; and generating a diagnosis result of the patient to be diagnosed according to the image data packet and the diagnosis model.
2. 2. The deep learning-based ovarian cancer diagnostic method of claim 1, wherein building a diagnostic model from historical diagnostic data of an ovarian cancer patient comprises: Determining a plurality of historical diagnosis strategies in the ovarian cancer full diagnosis period according to the historical diagnosis data of different ovarian cancer patients; generating a plurality of diagnosis sub-models according to each historical diagnosis strategy and corresponding historical diagnosis data, wherein each diagnosis sub-model corresponds to one historical diagnosis strategy; generating a plurality of characteristic indexes according to the historical basic parameters, and establishing a plurality of patient categories according to all the characteristic indexes; generating a preferred diagnostic strategy for each patient category; and constructing a diagnosis model according to the optimized diagnosis strategies of all the diagnosis sub-models and all the patient categories.
3. 3. The deep learning-based ovarian cancer diagnostic method of claim 2, wherein generating a preferred diagnostic strategy for each patient category comprises: Randomly selecting one patient category as a target patient category; generating a plurality of diagnosis strategies to be evaluated according to a plurality of historical diagnosis strategies corresponding to all the diagnosis sub-models; calculating the diagnosis evaluation value of each diagnosis strategy to be evaluated for the target patient class; Sequencing all diagnostic strategies to be evaluated according to the diagnostic evaluation values, and setting the first diagnostic strategy to be evaluated as a preferred diagnostic strategy of the target patient class; the preferred diagnostic strategy for each patient category is generated in turn.
4. 4. The deep learning-based ovarian cancer diagnostic method of claim 3, wherein calculating a diagnostic score for each diagnostic strategy to be evaluated for a target patient class comprises: presetting a plurality of diagnosis labels of a target patient category, wherein each diagnosis label is mapped with a corresponding weight coefficient; acquiring corresponding diagnosis data to be evaluated according to a plurality of historical diagnosis strategies corresponding to each diagnosis strategy to be evaluated, and inputting the diagnosis data to be evaluated into a corresponding diagnosis sub-model to obtain a corresponding diagnosis sub-label; generating a diagnosis label to be evaluated corresponding to the diagnosis strategy to be evaluated according to all diagnosis sub-labels of the same diagnosis strategy to be evaluated, performing comparative analysis with the corresponding diagnosis label, and generating a diagnosis sub-evaluation value of the corresponding diagnosis strategy to be evaluated for the corresponding diagnosis label according to an analysis result; And generating a diagnosis evaluation value of the corresponding diagnosis strategy to be evaluated for the target patient category according to the diagnosis sub evaluation value of the same diagnosis strategy to be evaluated for each diagnosis tag and the weight coefficient of each diagnosis tag.
5. 5. The deep learning based ovarian cancer diagnostic method of claim 4 wherein generating a diagnostic strategy for the patient to be diagnosed comprises: acquiring actual basic parameters of a patient to be diagnosed, and generating a plurality of actual characteristic indexes; performing similarity analysis on a plurality of actual characteristic indexes and characteristic indexes related to each patient category to obtain the similarity between the patient to be diagnosed and each patient category; the calculation formula of the similarity is as follows: ; wherein D is the similarity, n is the number of the actual characteristic indexes, ti is the similarity quantification value of the ith actual characteristic index and the corresponding characteristic index related to the patient category, A preset similarity quantization value threshold corresponding to an ith actual characteristic index, wherein ai is a weight coefficient of the ith actual characteristic index; sorting all patient categories according to the similarity, setting the first patient category as the patient category corresponding to the patient to be diagnosed, and setting the optimal diagnosis strategy of the corresponding patient category as the current diagnosis strategy of the patient to be diagnosed; and acquiring real-time image data according to a diagnosis strategy, constructing an image data packet, and inputting the image data packet into a diagnosis model to obtain a diagnosis result of a patient to be diagnosed.
6. 6. The deep learning-based ovarian cancer diagnostic method of claim 5, wherein generating a diagnostic result of the patient to be diagnosed from the image data package and the diagnostic model comprises: Inputting the image data packet into a diagnosis sub-model corresponding to a diagnosis strategy of a patient to be diagnosed to obtain a plurality of first diagnosis sub-labels; classifying the first diagnosis sub-label to obtain a normal label, a benign label, a malignant label and an unknown label; combining the normal labels to obtain a comprehensive normal label, comparing and analyzing the comprehensive normal label with a preset normal label, and generating a normal coefficient according to an analysis result; Combining the benign labels to obtain comprehensive benign labels, comparing and analyzing the comprehensive benign labels with a pre-constructed benign probability-benign label mapping table, and generating benign coefficients according to analysis results; combining the malignancy labels to obtain comprehensive malignancy labels, comparing and analyzing the comprehensive malignancy labels with a pre-constructed malignancy probability-malignancy label mapping table, and generating malignancy coefficients according to analysis results; Respectively calculating a first compensation coefficient, a second compensation coefficient and a third compensation coefficient according to the number of normal labels, the number of benign labels, the number of malignant labels and the number of unknown labels; Generating a comprehensive normal coefficient according to the first compensation coefficient and the normal coefficient; generating a comprehensive benign coefficient according to the second compensation coefficient and the benign coefficient; generating a comprehensive malignancy coefficient according to the third compensation coefficient and the malignancy coefficient; And generating a diagnosis result according to the comprehensive normal coefficient, the comprehensive benign coefficient and the comprehensive malignant coefficient.
7. 7. The deep learning-based ovarian cancer diagnostic method of claim 6, wherein generating a diagnostic result from the integrated normal coefficient, the integrated benign coefficient, and the integrated malignant coefficient comprises: presetting a normal coefficient threshold value, a benign coefficient threshold value and a malignant coefficient threshold value; if the comprehensive normal coefficient is larger than the normal coefficient threshold, the comprehensive benign coefficient is not larger than the benign coefficient threshold and the comprehensive malignant coefficient is not larger than the malignant coefficient threshold, generating a diagnosis result of the patient to be diagnosed according to the comprehensive normal label; If the comprehensive normal coefficient is not greater than the normal coefficient threshold, the comprehensive benign coefficient is greater than the benign coefficient threshold and the comprehensive malignant coefficient is not greater than the malignant coefficient threshold, generating a diagnosis result of the patient to be diagnosed according to the comprehensive benign label; If the comprehensive normal coefficient is not greater than the normal coefficient threshold, the comprehensive benign coefficient is not greater than the benign coefficient threshold and the comprehensive malignant coefficient is greater than the malignant coefficient threshold, generating a diagnosis result of the patient to be diagnosed according to the comprehensive malignant label; and if the comprehensive normal coefficient is not greater than the normal coefficient threshold, the comprehensive benign coefficient is not greater than the benign coefficient threshold and the comprehensive malignant coefficient is not greater than the malignant coefficient threshold, adjusting the unknown label and the diagnosis model, and reconstructing the diagnosis result.
8. 8. The deep learning-based ovarian cancer diagnosis method according to claim 7, wherein the calculating of the first compensation coefficient, the second compensation coefficient, and the third compensation coefficient according to the number of normal labels, the number of benign labels, the number of malignant labels, and the number of unknown labels, respectively, comprises: ; ; ; Wherein c1 is a first compensation coefficient, c2 is a second compensation coefficient, c3 is a third compensation coefficient, m1 is a normal label number, m2 is a benign label number, m3 is a malignant label number, m4 is an unknown label number, q1 is a first weight coefficient, q2 is a second weight coefficient, z1 is a first compensation conversion coefficient, z2 is a second compensation conversion coefficient, G1s is a predicted benign probability corresponding to an s-th benign label, G1 is a preset benign probability threshold, G2v is a predicted malignant probability corresponding to a v-th malignant label, ds is a weight coefficient of an s-th benign label, dv is a weight coefficient of a v-th malignant label.
9. 9. The deep learning-based ovarian cancer diagnostic method of claim 8, further comprising: Generating a visual report of the diagnosis result and feeding back the visual report to a user terminal, and setting a plurality of feedback time nodes corresponding to the patient; constructing a feedback data packet at a feedback time node; and obtaining the condition change condition of the corresponding patient according to the feedback data packet, dynamically evaluating the diagnosis result, and updating the diagnosis result and the diagnosis model according to the evaluation result.
10. 10. An ovarian cancer diagnostic system based on deep learning, comprising: The establishing module is used for establishing a diagnosis model according to historical diagnosis data of the ovarian cancer patient; The generation module is used for generating a diagnosis strategy of the patient to be diagnosed and acquiring an image data packet of the patient to be diagnosed according to the diagnosis strategy; And the diagnosis module is used for generating a diagnosis result of the patient to be diagnosed according to the image data packet and the diagnosis model.

Description

Deep learning-based ovarian cancer diagnosis method and system Technical Field The application relates to the technical field of ovarian cancer diagnosis, in particular to an ovarian cancer diagnosis method and system based on deep learning. Background Ovarian cancer is one of the common malignant tumors of female reproductive system, and has high morbidity and mortality, thereby seriously threatening the life health of females. In the prior art, the traditional ovarian cancer diagnosis method mainly depends on clinical experience of doctors, imaging examination and the like, however, individual differences of different patients and limitations of imaging examination are difficult to rapidly and accurately give diagnosis results, and the diagnosis results have larger errors, so that treatment time is easily delayed, and clinical treatment effects are reduced. Disclosure of Invention In order to solve the technical problems, the application provides an ovarian cancer diagnosis method and system based on deep learning, which are characterized in that a diagnosis model is built, corresponding diagnosis strategies are formulated for each patient to be diagnosed, an image data packet is acquired according to the diagnosis strategies and is input into the diagnosis model to obtain diagnosis results, individual differences of different patients and various historical diagnosis strategies in the whole diagnosis period of ovarian cancer can be comprehensively considered, so that a more accurate and comprehensive diagnosis model is built, the comprehensive diagnosis results are output and are subjected to dynamic evaluation, and the diagnosis results and the diagnosis model are updated according to the evaluation results, so that the accuracy and reliability of diagnosis are continuously improved, and the clinical treatment effect of ovarian cancer is effectively improved. In some embodiments of the present application, there is provided a deep learning-based ovarian cancer diagnosis method comprising: establishing a diagnosis model according to historical diagnosis data of the ovarian cancer patient; Generating a diagnosis strategy of a patient to be diagnosed, and acquiring an image data packet of the patient to be diagnosed according to the diagnosis strategy; and generating a diagnosis result of the patient to be diagnosed according to the image data packet and the diagnosis model. In some embodiments of the application, establishing a diagnostic model based on historical diagnostic data of an ovarian cancer patient includes: Determining a plurality of historical diagnosis strategies in the ovarian cancer full diagnosis period according to the historical diagnosis data of different ovarian cancer patients; generating a plurality of diagnosis sub-models according to each historical diagnosis strategy and corresponding historical diagnosis data, wherein each diagnosis sub-model corresponds to one historical diagnosis strategy; generating a plurality of characteristic indexes according to the historical basic parameters, and establishing a plurality of patient categories according to all the characteristic indexes; generating a preferred diagnostic strategy for each patient category; and constructing a diagnosis model according to the optimized diagnosis strategies of all the diagnosis sub-models and all the patient categories. In some embodiments of the application, generating a preferred diagnostic strategy for each patient category includes: Randomly selecting one patient category as a target patient category; generating a plurality of diagnosis strategies to be evaluated according to a plurality of historical diagnosis strategies corresponding to all the diagnosis sub-models; calculating the diagnosis evaluation value of each diagnosis strategy to be evaluated for the target patient class; Sequencing all diagnostic strategies to be evaluated according to the diagnostic evaluation values, and setting the first diagnostic strategy to be evaluated as a preferred diagnostic strategy of the target patient class; the preferred diagnostic strategy for each patient category is generated in turn. In some embodiments of the present application, calculating a diagnostic score for each diagnostic strategy to be evaluated for a target patient class includes: presetting a plurality of diagnosis labels of a target patient category, wherein each diagnosis label is mapped with a corresponding weight coefficient; acquiring corresponding diagnosis data to be evaluated according to a plurality of historical diagnosis strategies corresponding to each diagnosis strategy to be evaluated, and inputting the diagnosis data to be evaluated into a corresponding diagnosis sub-model to obtain a corresponding diagnosis sub-label; generating a diagnosis label to be evaluated corresponding to the diagnosis strategy to be evaluated according to all diagnosis sub-labels of the same diagnosis strategy to be evaluated, performing comparative analys