Dermatoscope image recognition method and device

Abstract

The invention provides a dermatoscope image recognition method and device, and the method comprises the steps: obtaining two dermatoscope images, which are obtained by enabling a same testee to receive dermatoscope examination for melanoma at different time, and contain a same interested target; obtaining a difference image according to the two dermatoscope images; identifying the two dermatoscopeimages and the difference image by using a machine learning model, wherein the model comprises two neural networks, the first neural network is used for carrying out feature extraction on the two dermatoscope images to obtain a first recognition result and determining feature difference data in the feature extraction process, and the second neural network is used for obtaining a second recognition result according to the difference image and the feature difference data; and determining whether the target of interest is an identification result of melanoma according to the first identificationresult and the second identification result.

Description

technical field [0001] The invention relates to the field of medical image detection, in particular to a dermoscopic image recognition method and equipment. Background technique [0002] Melanoma is the deadliest form of skin cancer. In the process of clinical diagnosis, doctors usually use a dermatoscope to conduct image inspections of skin lesions, and judge whether they belong to melanoma by observing structural characteristics such as shape, color, and size, and then give corresponding treatment plans. [0003] At present, the clinically common dermoscopy standardized diagnosis method is "ABCD criterion" or "ABCDE criterion", such as figure 1 As shown, A: shape symmetry; B: boundary irregularity; C: color type; D: diameter size; E: growth change degree. The internal mechanism is that benign lesions usually have a regular shape, relatively uniform color distribution, and a certain size within a certain range, and the overall characteristic state is stable; while melanom...

AI Technical Summary

Problems solved by technology

However, manual inspection has many limitations, its diagnostic accuracy is susceptible to subjective bias, and relies heavily on the experience of clinical practitioners

[0004] Image recognition schemes based on computer algorithms can alleviate the limitations of manual inspection to a certain extent. However, the current automatic melanoma recognition algorithm only relies on a single image of the lesion for recognition, ignoring the changes of the lesion.

figure 2 A comparison of the temporal growth stability of benign lesions and melanoma is given, where t1-t5 represents the different shooting times of dermoscopic images. In this case, it is difficult for the existing technology to give correct results for borderline suspected malignant lesions. The accuracy of mirror image recognition schemes still needs to be improved

Images