Optimizing time prediction and error classification in early melanoma detection using a hybrid RCNN-LSTM model.

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Melanoma skin cancer rates are increasing, making early detection critical.
• Deep learning algorithms have enhanced the accuracy of skin cancer predictive categorization.
• Malignant melanoma arises from abnormal melanocyte growth and requires timely diagnosis.

Purpose of the Study:

• To develop and evaluate a deep learning model for accurate and efficient early detection of malignant melanoma.
• To improve the classification accuracy and reduce prediction time for skin cancer detection using artificial intelligence.
• To analyze skin disease images for better understanding and prediction of melanoma.

Main Methods:

• Utilized the International Skin Image Collection dataset.
• Employed a Recurrent Convolutional Neural Network-Long Short-Term Memory (RCNN-LSTM) model for classification.
• Incorporated data preprocessing, feature extraction with RCNN, and classification with LSTM, considering context dependency and image augmentation.

Main Results:

• Achieved high performance metrics: 94.60% precision, 95.67% sensitivity, and 95.13% F1-score.
• Demonstrated reduced categorization error of 5.11% and overall accuracy of 95.42%.
• Reported shorter prediction durations (e.g., 95.314s for input size 10) and lower model loss (e.g., 0.15% for input size 20).

Conclusions:

• The RCNN-LSTM approach offers superior outcomes compared to previous methods for early melanoma detection.
• The model's consideration of context dependency prevents performance degradation and reduces classification errors.
• Image augmentation strategies minimize model loss, enhancing accuracy on unseen data for improved melanoma detection.