Automatic Detection and Segmentation of Colorectal Cancer with Deep Residual Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early detection of colorectal tumors is crucial for effective cancer treatment.
• Computer-aided diagnosis (CAD) systems utilize medical imaging, like CT scans, for tumor identification.
• Accurate segmentation and classification of colon lesions are necessary for optimal patient management.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (DCNN) for segmenting and classifying colorectal tumors.
• To improve the accuracy of tumor detection in CT images of the colon.
• To leverage residual network architecture for enhanced segmentation of colon cancer.

Main Methods:

• A two-phase approach involving segmentation followed by feature extraction and classification of colon lesions.
• Application of a deep convolutional neural network (DCNN) with a residual network (ResNet) architecture for segmenting polyps and the colon from 2D CT images.
• Incorporation of residual stack blocks with short skip connections to preserve spatial information during segmentation.

Main Results:

• The ResNet-enabled CNN model achieved effective segmentation of colorectal tumors.
• Performance metrics demonstrated high accuracy: 91.57% average Dice score, 98.28% sensitivity, 98.68% specificity, and 98.82% accuracy.
• The segmentation results served as features for successful classification of benign and malignant colon cancer.

Conclusions:

• The proposed DCNN model, utilizing a ResNet architecture, significantly enhances the segmentation and classification of colorectal tumors from CT images.
• This approach holds promise for improving early and automatic detection in computer-aided diagnosis (CAD) systems.
• The model's high performance metrics indicate its potential for clinical application in colorectal cancer analysis.