A Small Target Segmentation-Based Assistive Medical Diagnosis Method via Multimedia Data Analysis

The assistive medical diagnosis via multimedia data analysis has been a hot research topic in the area of intelligent health management, which saves a lot of human labor for the hospital. In that, a typical task is to detect and extract key small targets from the medical images. How to ensure detection speed and recognition precision is of great importance to the final practicability of the intelligent techniques. In this paper, we proposed a small target segmentation-based intelligent assistive diagnosis method for this purpose. In the stage of data pre-processing, data enhancement is employed to reduce the effects caused by noise, blur, and low image contrast. In the stage of decoding, a new upsampling method is designed to compensate for information in space and channels to help the network recover feature map resolution, reduce information loss, and improve the segmentation accuracy of smaller tissues or lesions. Such design of a multi-scale adaptive detail feature fusion module is able to make full use of features at different scales to recover high-level detail features, thus improving the segmentation accuracy of structurally complex tissues. Some simulative experiments aided by deep learning programming tools are conducted on the real-world multimedia data (CT, MRI, PET, etc.), so as to evaluate the proposed assistive diagnosis method. It can be concluded from the achieved results that the proposal is well suitable for multimedia data analysis of the experimental scenes.