Advanced deep learning for multi-class colorectal cancer histopathology: integrating transfer learning and ensemble...

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Colorectal cancer poses a significant global health challenge.
• Deep learning shows promise in improving early detection rates for colorectal cancer.
• Accurate classification of histopathological images is crucial for effective diagnosis.

Purpose of the Study:

• To optimize deep learning models for classifying colorectal cancer histopathology images.
• To enhance diagnostic accuracy for pathologists.
• To reduce colorectal cancer incidence and mortality through improved early detection.

Main Methods:

• Developed ensemble models using deep convolutional neural networks (CNNs).
• Applied data preprocessing including patch cropping, stain normalization, augmentation, and balancing.
• Utilized transfer learning for fine-tuning and pre-training CNN models.
• Evaluated performance on the Enteroscope Biopsy Histopathological Hematoxylin and Eosin Image (EBHI) dataset.

Main Results:

• Ensemble models achieved high classification accuracy across various magnifications (40x, 100x, 200x, 400x).
• Highest accuracy reached 99.36% on the 100x magnification subset.
• Demonstrated exceptional performance in recall, precision, and F1 score.

Conclusions:

• Ensemble deep learning models show strong performance for colorectal cancer histopathology image classification.
• Findings support the potential for improved early detection and accurate diagnosis of colorectal cancer.
• This approach can aid pathologists in precise diagnostic analysis.