Dual-Path Convolutional Neural Network with Squeeze-and-Excitation Attention for Lung and Colon Histopathology...

Area of Science:

• * Computational pathology
• * Artificial intelligence in oncology
• * Medical image analysis

Background:

• * Lung and colon cancers are leading causes of mortality, necessitating precise histopathological diagnosis.
• * Manual slide examination is time-consuming and subject to inter-observer variability.
• * Development of automated, explainable diagnostic systems is crucial for improving accuracy and efficiency.

Purpose of the Study:

• * To introduce DPCSE-Net, a lightweight dual-path convolutional neural network with a squeeze-and-excitation (SE) attention mechanism.
• * To evaluate DPCSE-Net's performance in classifying lung and colon cancer histopathology.
• * To enhance model interpretability using visualization techniques like Grad-CAM and Integrated Gradients.

Main Methods:

• * Developed a dual-path convolutional neural network (DPCSE-Net) integrating multiscale feature extraction and SE attention.
• * Employed Gradient-weighted Class Activation Mapping (Grad-CAM), attention heatmaps, and Integrated Gradients for model interpretability.
• * Validated DPCSE-Net on the LC25000 dataset for lung and colon cancer classification.

Main Results:

• * DPCSE-Net achieved state-of-the-art accuracy (99.88%) and F1-score on the LC25000 dataset.
• * The model demonstrated high efficiency with low computational complexity.
• * Ablation studies confirmed the effectiveness of the dual-path design and SE module.
• * Qualitative analysis showed that the model's focus aligns with diagnostically relevant regions.

Conclusions:

• * DPCSE-Net provides an accurate, efficient, and explainable framework for computer-aided histopathological diagnosis.
• * The study supports the integration of explainable AI in medical image analysis for cancer detection.
• * DPCSE-Net has the potential to assist pathologists in making faster and more reliable diagnoses.