MedNet: a lightweight attention-augmented CNN for medical image classification.

Area of Science:

• Medical imaging analysis
• Computer vision
• Deep learning for healthcare

Background:

• Accurate medical image classification is crucial for early disease diagnosis.
• Standard deep learning models struggle with medical image complexities like low resolution and high intra-class variability.
• Existing models often lack efficiency and effective feature extraction for medical data.

Purpose of the Study:

• To develop an efficient and accurate deep learning architecture for medical image classification.
• To address the challenges of low inter-class variance and high intra-class variability in medical images.
• To propose MedNet, a lightweight CNN incorporating depthwise separable convolutions and CBAM attention.

Main Methods:

• Proposed MedNet, a lightweight Convolutional Neural Network (CNN) architecture.
• Integrated depthwise separable convolutions for efficient feature extraction.
• Employed CBAM (Convolutional Block Attention Module) for refining spatial and channel-wise features.
• Utilized adaptive pooling, dropout, and fully connected layers for classification.
• Trained and validated on DermaMNIST, BloodMNIST, OCTMNIST (MedMNIST), and Fitzpatrick17k datasets.

Main Results:

• MedNet achieved performance matching or exceeding established CNN baselines across multiple medical image datasets.
• Demonstrated higher accuracy compared to existing models.
• Achieved significantly fewer model parameters and lower computational cost.
• Showcased effective extraction and refinement of critical spatial and contextual features.

Conclusions:

• MedNet offers an effective and efficient solution for medical image classification.
• The proposed architecture successfully mitigates challenges associated with medical image data.
• MedNet presents a promising lightweight model for reliable medical image analysis and diagnosis.