Deep learning-based classification of colonoscopic images using an attention-enhanced ConvNeXt V2 architecture

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate interpretation of colonoscopic images is crucial for early detection of colorectal diseases.
• Automated classification of these images is challenging due to visual complexity and similarity.

Purpose of the Study:

• To develop an attention-enhanced deep learning framework for robust multi-class classification of colonoscopic images.
• To improve diagnostic performance in identifying conditions such as polyps and ulcerative colitis.

Main Methods:

• Proposed an attention-enhanced deep learning framework utilizing ConvNeXt V2 architecture.
• Integrated a Convolutional Block Attention Module (CBAM) to focus on diagnostically relevant regions.
• Trained and evaluated the model on a balanced dataset of cecum, polyp, and ulcerative colitis images using 5-fold cross-validation and data augmentation.

Main Results:

• Achieved a mean classification accuracy of approximately 95%, outperforming the baseline ConvNeXt V2 model (approx. 90%).
• Obtained a mean precision of 95.1% and an F1-score of 94.9%, indicating reliable classification across all classes.
• Attention visualization confirmed the model's ability to focus on clinically relevant pathological features.

Conclusions:

• Modern convolutional architectures with attention mechanisms enhance diagnostic performance in colonoscopic image analysis.
• The proposed framework offers an efficient tool for automatic classification of colorectal diseases, aiding clinical decision-making.