Texture based skin lesion abruptness quantification to detect malignancy.

Area of Science:

• Dermatology
• Medical Imaging
• Computational Pathology

Background:

• The abruptness of pigment patterns at a skin lesion's periphery is crucial for detecting malignancy.
• Current clinical methods for assessing lesion border abruptness are subjective and prone to errors.

Purpose of the Study:

• To develop and validate an improved computational model for quantitatively measuring skin lesion border abruptness.
• To enhance the accuracy and objectivity of melanoma detection using dermoscopic images.

Main Methods:

• A computational model was developed to analyze texture features within a defined region at the lesion periphery.
• The Level Set Propagation (LSP) method was employed to determine an interior border line for analysis.
• Feature vectors including homogeneity, standard deviation, and mean pixel values were extracted and classified using neural networks (NN) and Support Vector Machine (SVM).

Main Results:

• The model achieved an 87% f1-score and 78% specificity in classifying melanoma cases.
• The LSP method demonstrated superior performance compared to previous studies.
• The best classification results were obtained using a fully-connected multi-hidden layer NN with RGB color spaces.

Conclusions:

• Computational analysis of skin lesion border abruptness is a reliable indicator of malignancy.
• The developed computational model effectively quantifies lesion border abruptness using texture homogeneity and LSP, offering an objective diagnostic tool.