Multifunctional Breast Cancer Diagnosis And Analysis Platform Based On Artificial Intelligence

Breast cancer is the second-ranked disease in female cancer mortality.Early detection and diagnosis are the key to controlling the disease.Ultrasound imaging has the advantages of high cost performance and simple operation,and is widely used in the diagnosis of breast cancer.At present,artificial intelligence has made great progress in natural images and other fields,but there are few studies in the field of breast ultrasound imaging.The application of artificial intelligence technology to the diagnosis of breast cancer is a good research problem.This article mainly studies the multi-functional breast cancer diagnosis based on artificial intelligence.The breast ultrasound image data set used is from Harbin Medical University Tumor Hospital and confirmed by biopsy.We first analyze the historical research and current status of breast cancer diagnosis,and then give a detailed introduction on how to establish a large-scale ultrasound data set,and perform benign and malignant diagnosis of breast tumors,molecular classification prediction,and BI-RADS based on the established data set.Research on hierarchical tasks,and use interpretable deep learning methods for task diagnosis models to analyze the differences between the models.Finally,the above-mentioned algorithms are integrated into the Web platform and combined with the needs of actual task diagnosis,a multi-functional breast cancer diagnosis and analysis platform based on manual is built.In general,the main research contents and contributions of this article are as follows: In summary,this paper has the following contributions:1.This paper established a large-scale high-quality ultrasound image data set.It contains3049 breast ultrasound images of 2092 tumor patients.According to the research,this data set is the largest known breast ultrasound image data set in China.And the data set contains a lot of useful medical history information about the patient,such as molecular typing,pathological information,BI-RADS tags,etc.,and image restoration is performed through the designed image restoration software,which is not available in other ultrasound image datasets.2.This paper adopts the transfer learning method to diagnose benign and malignant breast cancer.The diagnostic accuracy reaches 87.75%,which is 3% higher than the deep learning model of non-transfer learning,and the Xception model suitable for diagnosis is selected as the breast Cancer diagnosis task model.3.This paper innovatively uses the method of multiple migration learning to predict molecular typing,which solves the problem of small-scale data analysis of breast cancer molecular typing.This is the first time that artificial intelligence methods have been used to predict the molecular classification of breast ultrasound images,and the results have achieved AUC values above 0.8,which fully demonstrates the strong correlation between molecular classification and tumor images.This research is important for the future the clinical diagnosis of breast cancer is of great significance.4.In the breast cancer diagnosis method evaluated by BI-RADS,innovatively add the doctor’s prior knowledge to the training of breast cancer diagnosis model.The designed diagnosis accuracy of breast cancer diagnosis model based on multi-task migration learning reached 90.7%,which was increased by 2.95% compared with the single-task model.The model incorporates the diagnosis of the BI-RADS evaluation system,and the AUC value of the BI-RADS classification reaches 0.8352,which is closer to the doctor’s evaluation standard.The accuracy of the combined diagnosis reached 93.38%,and the AUC value was 0.9820.The performance was greatly improved,which further met the requirements of clinical diagnosis.5.In the interpretability model analysis,this article analyzes the improvement of model diagnosis ability by transfer learning,multiple transfer learning,and multi-task learning.The model diagnosis area and characteristic characteristics are described through the interpretable heat map.Corresponding analysis has been carried out for the improvement of,which provides a model interpretation method for the application in actual diagnosis scenarios.6.At the end of this paper,the algorithms of image processing and deep learning diagnosis model in the previous chapters are embedded into the designed web-side application,combined with the actual needs of breast cancer diagnosis at present,a multifunctional breast cancer diagnosis based on artificial intelligence is constructed.Analysis platform.The platform can perform multi-functional diagnosis of breast cancer,markup of high-quality breast cancer image data,and upload medical records.As far as we know,this is the most complete and systematic Web-side breast cancer diagnostic analysis platform established at home and abroad.