Early diagnosis for cutaneous malignant melanoma based on the intellectualized classification and recognition for images of melanocytic tumour by dermoscopy

Abstract

Background: In recent years, the morbidity of Melanocytic Tumor, especially Cutaneous Malignant Melanoma, has been increasing year by year. Cutaneous Malignant Melanoma has been the most fatal skin disease owing to its high malignant level and proneness to metastasis. It is of great importance to establish the intellectualized classification and recognition for Melanocytic Tumor in the yellow race, in order to realize the early diagnosis and reduce the mortality for Cutaneous Malignant Melanoma.

Methods: We adopted the polarized-light dermoscopy image technology; acquired the images of Melanocytic Tumor from the yellow race in a non-invasive way; proposed the algorithms of adaptive clustering segmentation and feature extraction; quantificationally analyzed the six features of Melanocytic Tumor images, including its asymmetry, eccentricity, border depressed rate, uniformity of radiation in transition area, color diversity and texture correlation; realized the classification and intellectualized recognition for benign or malignant Melanocytic Tumor, with the combined neural network classifier; verified the results by histopathology and statistical analysis.

Results: Among the 642 images, the benign ones accounted for 82.4%, the malignant ones 17.6%. The sensitivity and the specificity of uniformity of radiation in transition area, texture correlation, border depressed rate and color diversity were from 86.73% to 95.58%, and from 97.3% to 100% respectively, which showed relatively high performance. The sensitivity and the specificity of asymmetry and eccentricity were from 41.59% to 47.78%, and from 69.91% to 76.99% respectively, which showed relatively low performance. The average classification accuracy for benign and malignant Melanocytic Tumor reached 93.65% by the multiple independent neural network classifiers, the difference of which was significant by chi-square test (χ² = 4.51, P＜0.05).

Conclusions: The polarized-light dermoscopy analysis technology is able to realize the feature classification and automatic recognition for benign or malignant Melanocytic Tumor in a non-invasive way. It will lay foundations for solving the key problem of early diagnosis and intellectualized recognition for Cutaneous Malignant Melanoma in the yellow race.