EnsemDeepCADx: Empowering Colorectal Cancer Diagnosis with Mixed-Dataset Features and Ensemble Fusion CNNs on...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Colorectal cancer presents a significant mortality risk, necessitating accurate and timely diagnosis.
• Colonoscopy images are crucial for colorectal cancer diagnosis, making diagnostic accuracy paramount.
• Deep learning techniques offer potential to enhance the accuracy of existing diagnostic systems.

Purpose of the Study:

• To develop a novel deep learning system, EnsemDeepCADx, for accurate colorectal cancer diagnosis.
• To leverage advanced deep learning techniques, including Convolutional Neural Networks (CNNs) and transfer learning, to improve diagnostic performance.

Main Methods:

• Developed the EnsemDeepCADx system by combining CNNs with transfer learning using bidirectional long short-term memory (BILSTM) and support vector machines (SVM).
• Utilized four pre-trained CNN models (AlexNet, DarkNet-19, DenseNet-201, ResNet-50) within ensemble CNNs (ADaDR-22, ADaR-22, DaRD-22).
• Evaluated the system in stages using diverse datasets (colour, greyscale, LBP, feature fusion) from CKHK-22, incorporating feature fusion and transfer learning techniques.

Main Results:

• The EnsemDeepCADx system achieved optimal testing accuracy of 95.96% on the original dataset, 88.79% on greyscale, 73.54% on LBP, and 97.89% on the feature fusion dataset.
• The ensemble fusion CNN DaRD-22 combined with BILSTM and SVM demonstrated superior performance across all tested datasets.
• The system's accuracy was maximized by comparing outputs across multiple feature datasets and ensemble CNNs at each evaluation stage.

Conclusions:

• The EnsemDeepCADx system, utilizing a combination of ensemble CNNs, BILSTM, and SVM, significantly enhances diagnostic accuracy for colorectal cancer.
• The proposed deep learning approach demonstrates a highly effective method for improving colorectal cancer detection from colonoscopy images.
• The study highlights the potential of advanced AI techniques in improving patient outcomes for high-mortality diseases like colorectal cancer.