[Paper Review] Segmentation and ABCD rule extraction for skin tumors classification

Mahammed Messadi, Hocine Cherifi+1 more|arXiv (Cornell University)|Jun 8, 2021

Cutaneous Melanoma Detection and Management36 references22 citations

TL;DR

This paper proposes an automated skin tumor classification system using the clinical ABCD rule (Asymmetry, Border, Color, Diversity) for melanoma detection. It combines morphological preprocessing, unsupervised level set segmentation initialized via iterative thresholding, and a neural network classifier, achieving improved true detection and reduced false positives on a 320-image dermoscopic dataset.

ABSTRACT

During the last years, computer vision-based diagnosis systems have been widely used in several hospitals and dermatology clinics, aiming at the early detection of malignant melanoma tumor, which is among the most frequent types of skin cancer. In this work, we present an automated diagnosis system based on the ABCD rule used in clinical diagnosis in order to discriminate benign from malignant skin lesions. First, to reduce the influence of small structures, a preprocessing step based on morphological and fast marching schemes is used. In the second step, an unsupervised approach for lesion segmentation is proposed. Iterative thresholding is applied to initialize level set automatically. As the detection of an automated border is an important step for the correctness of subsequent phases in the computerized melanoma recognition systems, we compare its accuracy with growcut and mean shift algorithms, and discuss how these results may influence in the following steps: the feature extraction and the final lesion classification. Relying on visual diagnosis four features: Asymmetry (A), Border (B), Color (C) and Diversity (D) are computed and used to construct a classification module based on artificial neural network for the recognition of malignant melanoma. This framework has been tested on a dermoscopic database [16] of 320 images. The classification results show an increasing true detection rate and a decreasing false positive rate.

Motivation & Objective

• To develop an automated system for early detection of malignant melanoma using computer vision.
• To reduce the impact of small anatomical structures on lesion segmentation through morphological and fast marching preprocessing.
• To improve segmentation accuracy by comparing level set, growcut, and mean shift algorithms for lesion border detection.
• To extract clinical ABCD features (Asymmetry, Border, Color, Diversity) for use in a neural network-based classification model.
• To evaluate the system's performance on a standardized dermoscopic image database for melanoma classification.

Proposed method

• Apply morphological operations and fast marching schemes to preprocess dermoscopic images and suppress noise and small structures.
• Use iterative thresholding to automatically initialize the level set method for unsupervised lesion segmentation.
• Compare segmentation performance of level set, growcut, and mean shift algorithms to select the most accurate method.
• Extract four visual diagnostic features—Asymmetry (A), Border (B), Color (C), and Diversity (D)—from segmented lesions.
• Train an artificial neural network classifier using the extracted ABCD features to distinguish benign from malignant skin tumors.
• Validate the system on a public dermoscopic dataset of 320 images to assess classification performance.

Experimental results

Research questions

• RQ1 How does preprocessing with morphological and fast marching methods affect lesion segmentation accuracy?
• RQ2 Which segmentation technique—level set, growcut, or mean shift—yields the most accurate lesion boundary detection for melanoma classification?
• RQ3 To what extent do ABCD rule-based features improve the performance of a neural network in classifying skin tumors?
• RQ4 Can an automated system based on clinical ABCD rules achieve high true detection and low false positive rates in melanoma diagnosis?
• RQ5 How does the integration of automated segmentation and feature extraction impact overall classification performance on real dermoscopic images?

Key findings

• The proposed preprocessing step effectively reduces the influence of small structures, improving segmentation stability.
• The level set method with iterative thresholding initialization achieved superior segmentation accuracy compared to growcut and mean shift.
• The ABCD-based feature extraction pipeline successfully captured clinically relevant characteristics of skin lesions.
• The neural network classifier demonstrated an increasing true detection rate and decreasing false positive rate on the 320-image dataset.
• The system's performance was validated on a real-world dermoscopic database, confirming its potential for clinical deployment.
• The integration of automated segmentation and ABCD rule extraction enables a clinically interpretable and accurate melanoma classification system.