Medical image deep learning method with interpretability

Abstract

A medical image deep learning method with interpretability comprises the following steps: 1) acquiring an original medical image, performing image enhancement on the original medical image, increasing the number of the images, and inputting the images as input images into a network; 2) extracting image features of the input image by using a VGG-16 network to obtain a medical predictive map conforming to intermediate clinical features; 3) comparing the obtained prediction map with a standard intermediate clinical feature, calculating the cross entropy loss Lossx, y of the two, and adjusting VGG-16 network parameters to make the calculated cross entropy loss smaller than a set threshold so as to complete network learning and make the prediction map obtained by the VGG-16 network accord with the requirement of a comparison result. The network obtained through the method has better detection precision and better interpretability, and subsequent network optimization can be conveniently carried out according to requirements.

Description

technical field [0001] The invention relates to the field of computer images, in particular to an explainable deep learning method for medical images. Background technique [0002] In the diagnosis process of glandular cancer, it is very important to be able to automatically identify pathological regions and divide them accurately. In the field of traditional medicine, this process is generally done manually. However, due to the large number of sliced ​​images, the wide coverage of glands, and the various manifestations due to different physiological characteristics of patients, inexperienced doctors often miss or misdiagnose. Therefore, it is important to be able to perform accurate glandular cancer detection and case division in advance. Among them, the deep learning neural network for accurate division of medical gland image cancer detection can determine the current stage of the disease to a large extent, and help doctors make reasonable diagnosis and treatment. [00...

AI Technical Summary

Problems solved by technology

At the same time, due to the protection of patient privacy and the extremely small proportion of images that can be used for glandular cancer detection and segmentation, the number of images that can be used for training and testing is extremely small; the incidence of different patients is different, and the boundary The pixels are fuzzy, easily confused with the surrounding normal physiological features, and the lesion area may be missed or misdiagnosed

Images