Lesion localization method for skin disease image

Abstract

The invention discloses a lesion localization method for a skin disease image. By training an identification model of skin diseases and a generation model of a lesion candidate frame, a picture containing a candidate frame is processed by using a template shielding method to realize lesion localization. Google inception v3 architecture and migration learning are used for training. An OpenCV cascading training classifier is used for training positive and negative samples to generate an xml document. For the generated lesion candidate frame, the template shielding method is adopted; each of thecandidate frames is identified and judged by the identification model one by one, thereby filtering out the real lesion. According to the lesion localization method for the skin disease image in the invention, in the case that a doctor does not need to label the picture during test, the obtained lesion localization model still can circle the lesion area accurately and quickly, thereby saving the time and effort for the doctor to mark a large number of pictures.

Description

technical field [0001] The invention relates to medical data processing, in particular to a lesion localization method for skin disease images. Background technique [0002] Generally speaking, lesion localization can be understood as the detection of targets. Nowadays, there are two main types of target detection and recognition research methods: (1) target detection and recognition methods based on traditional image processing and machine learning algorithms, traditional target detection The identification method can be mainly expressed as: target feature extraction -> target recognition -> target positioning. (2) Target detection and recognition method based on deep learning. With the development of deep learning, target detection and recognition based on deep learning has become the mainstream method, which can be mainly expressed as: deep feature extraction of images -> target recognition and positioning based on deep neural network, in which the deep neural n...

AI Technical Summary

Problems solved by technology

[0004] (1) The manual method of lesion localization is largely affected by human subjective consciousness, which may cause misjudgment of image information and affect subsequent pathological analysis

[0005] (2) In China, there is a shortage of medical personnel to varying degrees in both large urban hospitals and small rural hospitals, which means that professional dermatologists are required to circle a large number of lesion areas on skin disease images. Unrealistic, this will cause a serious waste of medical resources

[0006] (3) Due to the complexity and irregularity of skin diseases and the differences between different individuals, the general image segmentation method is not ideal for skin disease images, resulting in inaccurate positioning of the lesion area

[0007] (4) Although target detection algorithms such as yolo and Faster-R-CNN can perform better positioning in ordinary daily pictures, their training requires a large amount of data, and medical images are very different from daily pictures. Class target detection algorithm cannot accurately extract medical image features, which will lead to low accuracy of lesion location

Images