Skin Lesion Detection Using Hand-Crafted and DL-Based Features Fusion and LSTM

Area of Science:

• Dermatology
• Computational Pathology
• Medical Imaging Analysis

Background:

• Skin cancer diagnosis relies on time-consuming methods like imaging and biopsies.
• Existing automated methods using hand-crafted features may lack robustness for early-stage detection.
• Need for accurate and efficient automated skin cancer detection models.

Purpose of the Study:

• To propose a novel and robust skin cancer detection model based on feature fusion.
• To integrate both machine learning (ML) and deep learning (DL) approaches for improved performance.
• To accurately classify skin lesions as benign or malignant.

Main Methods:

• Image pre-processing using a Gaussian filter (GF) to reduce noise.
• Hybrid feature extraction combining Local Binary Patterns (LBP) and Inception V3.
• Feature fusion followed by classification using a Long Short-Term Memory (LSTM) network with an Adam optimizer.

Main Results:

• Achieved 99.4% accuracy, 98.7% precision, 98.66% recall, and 98% F-score on the DermIS dataset.
• Cross-validation on the International Skin Image Collection (ISIC) dataset yielded 98.4% detection accuracy.
• The proposed feature fusion method significantly outperformed existing segmentation-based and DL-based techniques.

Conclusions:

• The proposed feature fusion model demonstrates superior performance in skin cancer detection.
• Combining ML and DL features enhances the robustness and accuracy of the diagnostic model.
• This approach offers a promising automated solution for early and accurate skin cancer classification.