Colour and contrast enhancement for improved skin lesion segmentation

Area of Science:

• Dermatology
• Medical Image Analysis
• Computer Vision

Background:

• Accurate segmentation of skin lesions from dermoscopic images is essential for diagnosis.
• Challenges include poor image contrast and color variations due to differing illumination and devices.
• These issues hinder precise border detection and segmentation performance.

Purpose of the Study:

• To develop and evaluate a pre-processing technique to enhance color information and contrast in dermoscopic images.
• To improve the performance of existing lesion segmentation algorithms.
• To assess the impact of the proposed enhancement on segmentation accuracy.

Main Methods:

• A pre-processing step was designed to enhance image contrast and color information.
• The enhanced images were used with two segmentation algorithms: background analysis and co-operative neural networks for edge detection.
• Performance was evaluated on a dataset of 100 dermoscopic images with expert-verified ground truths.

Main Results:

• Both segmentation techniques, when applied after the enhancement step, achieved good performance.
• The color and contrast enhancement pre-processing significantly improved segmentation results compared to using original RGB images.
• The enhancement step was demonstrated to be crucial for accurate lesion border extraction.

Conclusions:

• The proposed pre-processing method effectively enhances dermoscopic images for improved skin lesion segmentation.
• Color and contrast enhancement are critical for robust performance of segmentation algorithms in medical imaging.
• This approach offers a valuable tool for more accurate analysis of dermoscopic images.