Accurate colorectal cancer detection using a random hinge exponential distribution coupled attention network on...

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Colorectal cancer (CRC) poses a significant global health challenge, with early detection crucial for improved patient outcomes.
• Traditional pathological image analysis for CRC is labor-intensive and susceptible to human error, highlighting the need for automated solutions.

Purpose of the Study:

• To develop and evaluate an advanced deep learning approach, the Random Hinge Exponential Distribution coupled Attention Network (RHED-CANet), for automated CRC detection.
• To enhance the accuracy and efficiency of colorectal cancer diagnosis using pathological images.

Main Methods:

• Utilized TCGA-CRC-DX and CRC datasets for training and validation.
• Employed a Modified Square Root Sage-Husa Adaptive Kalman Filter and Spike-Driven Transformer for pre-processing and feature extraction.
• Implemented an EfficientNetV2L Inception Transformer for precise segmentation of cancerous regions.
• Applied the RHED-CANet for final classification of colorectal cancer pathological images.

Main Results:

• Achieved a diagnostic accuracy of 99.9% and a precision of 99.7%.
• Demonstrated significant reduction in diagnostic time compared to traditional methods.
• The model effectively minimizes false positives, enhancing overall diagnostic reliability.

Conclusions:

• The proposed RHED-CANet method shows exceptional performance in detecting colorectal cancer from pathological images.
• Potential limitations include dataset overfitting, computational complexity, and potential bias affecting training efficiency for rare subtypes.
• Despite limitations, the approach represents a promising advancement for clinical applications in colorectal cancer diagnostics.