Diagnosis model and device for cervical atypical lesions based on multimodal attention model

Abstract

The invention discloses a cervical atypical lesion diagnosis model and device based on a multimodal attention model, which belongs to the field of medical artificial intelligence. Firstly, the patient's cervical acetic acid map and iodine map are obtained, and the disease level of the patient is divided into normal and LSIL. , HSIL three categories, use this category as the label corresponding to the acetic acid map and the iodine map to form training data. After data preprocessing, the multi-modal fusion model is passed in, and the multi-layer feature maps learned in the model are fused separately. When fused, the attention mechanism is introduced to select a mode with a good classification effect and generate The auxiliary attention information is applied to the feature map in the mode with poor effect, and the fusion operation is performed layer by layer, and the probability of the final output image belonging to the three categories is repeated. The above process is repeated to iteratively train the model until convergence. Afterwards, the images that need to be diagnosed with lesion categories are input into the trained model, and the above-mentioned feature fusion method is used to output the corresponding prediction results to assist doctors in diagnosis.

Description

technical field [0001] The invention relates to the field of medical artificial intelligence, in particular to a model and device for diagnosing cervical atypical lesions based on a multimodal attention model. Background technique [0002] Cervical cancer is currently the main disease that threatens women's health, and it is also the second leading cancer that causes women's death worldwide. It seriously destroys the sex life of patients and affects the quality of life. There are about 150,000 new cases and about 100,000 women die of cervical cancer in my country every year. [0003] By detecting cervical squamous intraepithelial lesion (cervical intraepithelial neoplasia, CIN), can help patients and doctors to prevent cervical cancer, in medicine, CIN can be divided into two groups: low-grade squamous intraepithelial lesion (low-grade squamous intraepithelial lesion) lesion, LSIL) and high-grade squamous intraepithelial lesion (high-gradesquamous intraepithelial lesion, HS...

AI Technical Summary

Problems solved by technology

However, this simple fusion method will destroy the features in the image, causing the features learned by the network to be confused. At the same time, the performance of this method is not stable enough, and it will significantly reduce the number of training samples.

Images