An attention based hybrid approach using CNN and BiLSTM for improved skin lesion classification

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Skin lesions pose a growing global health challenge, necessitating improved diagnostic accuracy for effective treatment.
• Early and precise detection of skin lesions is critical for enhancing patient outcomes and managing disease progression.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning model for accurate skin lesion classification.
• To improve the diagnostic capabilities for various skin lesions using advanced artificial intelligence techniques.

Main Methods:

• Integration of Convolutional Neural Networks (CNNs) with Bidirectional Long Short Term Memory (BiLSTM) networks.
• Enhancement of the hybrid model with spatial, channel, and temporal attention mechanisms for improved feature extraction.
• Comparative analysis against various deep learning architectures including standalone CNNs, InceptionV3, VGG16, and Xception.

Main Results:

• The proposed CNN-BiLSTM model with attention mechanisms achieved superior performance, with an accuracy of 92.73%.
• Key performance metrics included precision (92.84%), F1 score (92.70%), recall (92.73%), Jaccard Index (87.08%), Dice Coefficient (92.70%), and MCC (91.55%).
• The hybrid model outperformed other configurations, demonstrating the efficacy of the attention-enhanced approach.

Conclusions:

• The developed deep learning model offers a robust tool for healthcare professionals, enhancing the accuracy of skin lesion diagnosis.
• This advancement supports proactive management strategies and personalized treatment, potentially mitigating the global impact of skin lesions.
• The research contributes to technological advancements in medical diagnostics, aiming to improve public health resilience through timely interventions.