An efficient multi-level pre-processing algorithm for the enhancement of dermoscopy images in melanoma detection

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Automated skin lesion detection is crucial for early melanoma diagnosis.
• Dermoscopy images often suffer from artifacts like noise, uneven illumination, and hair, hindering accurate segmentation.
• Existing pre-processing methods may not adequately address these challenges.

Purpose of the Study:

• To propose a novel multi-level pre-processing algorithm for dermoscopy images.
• To enhance image quality for improved automated skin lesion segmentation.
• To evaluate the effectiveness of the proposed pre-processing steps on segmentation accuracy.

Main Methods:

• A sequence of pre-processing steps including de-noising (Non-Local Means filter), illumination correction (gamma correction), contrast enhancement (Robust Image Contrast Enhancement), sharpening (unsharp masking), reflection removal, and virtual shaving (phase congruency-based method).
• Utilized Blind Reference less Image Spatial Quality Evaluator (BRISQUE) for quality assessment.
• Employed Regularized Extreme Learning Machine for automated skin lesion segmentation.

Main Results:

• The proposed multi-level pre-processing significantly improved image quality, with superior performance over existing methods in contrast, information preservation, and artifact removal.
• Non-Local Means filter and phase congruency-based virtual shaving showed optimal results for de-noising and hair removal, respectively.
• Segmentation accuracy using Regularized Extreme Learning Machine was substantially enhanced when applied to pre-processed images.

Conclusions:

• The developed multi-level pre-processing algorithm effectively enhances dermoscopy images for skin lesion analysis.
• The sequence of pre-processing steps is vital for maximizing the performance of automated skin lesion segmentation.
• This approach offers a robust solution for improving the accuracy and reliability of melanoma detection systems.