Light Field Image Dataset of Skin Lesions.

Area of Science:

• Medical Imaging
• Computer Vision
• Dermatology

Background:

• Light field imaging captures rich visual data beyond traditional 2D systems, enabling dense multiview and depth information.
• Emerging applications in medical imaging, particularly dermatology, leverage light field technology for novel feature discovery and improved classification.
• Machine learning approaches show promise for analyzing complex medical image data.

Purpose of the Study:

• To introduce a publicly available light field image dataset of skin lesions (SKINL2 v1.0) for the research community.
• To facilitate advancements in medical imaging research and the development of new classification algorithms for skin lesions.
• To support clinically-oriented dermatology studies through enriched visual data.

Main Methods:

• Acquisition of 250 light fields using a focused plenoptic camera, with each light field containing 81 views of a skin lesion.
• Classification of lesions into eight clinical categories.
• Inclusion of dermatoscopic images for each lesion.
• Characterization of a subset of 17 central view images based on spatial information (SI), colourfulness (CF), and compressibility.

Main Results:

• The SKINL2 v1.0 dataset comprises 250 light fields of skin lesions, each with 81 views.
• The dataset is classified into eight clinical categories and includes dermatoscopic images.
• A subset of images was analyzed for spatial information, colourfulness, and compressibility.

Conclusions:

• The SKINL2 v1.0 dataset offers significant potential for advancing light field-based medical imaging and machine learning classification algorithms in dermatology.
• This resource can accelerate research in developing more accurate and robust diagnostic tools for skin conditions.
• The dataset supports both technical research in imaging and clinical studies in dermatology.