Image Processing & Autoencoder Boost Two-Stage Filtering For Pigmented Skin Lesion Detection

Abstract

It is critical to diagnose pigmented skin lesions, which might be benign or malignant, as early as possible. For better detection, this paper suggests a two-stage filtering method that is followed by majority voting. In stage 1 filtering Hough line transformation, morphological operations, and Canny edge detection are some of the image processing techniques used to remove hair. Stage 2 filtering uses a convolutional autoencoder to extract features and reduce noise even more. Through majority voting, combined predictions from deep learning (CNN) and machine learning (Random Forest) models, both trained on the massively filtered dataset (HAM10000), are blended to produce results that are more accurate than those of the individual models. This ensemble approach demonstrates promising potential for AI-powered primary care diagnosis of pigmented skin lesions, aiding early and precise identification.