Automatic Polyp Segmentation in Colonoscopy Images Using a Modified Deep Convolutional Encoder-Decoder Architecture.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Colorectal cancer is a leading cause of cancer death globally.
• Optical colonoscopy is standard for polyp detection but is operator-dependent.
• Automated polyp segmentation is needed for efficient colorectal cancer screening.

Purpose of the Study:

• To develop a computer-aided diagnosis (CAD) method for automatic polyp segmentation in colonoscopy images.
• To enhance the accuracy and efficiency of colorectal cancer diagnosis.

Main Methods:

• A modified SegNet VGG-19 convolutional neural network was proposed for polyp segmentation.
• Modifications included skip connections, 5x5 convolutional filters, and parallel dilated convolutions.
• The model was trained and evaluated on CVC-ClinicDB, CVC-ColonDB, and ETIS-LaribPolypDB datasets.

Main Results:

• The proposed model achieved high performance metrics: 96.06% accuracy, 94.55% sensitivity, 97.56% specificity, 97.48% precision, 92.3% mean IoU, and 95.99% dice coefficient.
• Performance was comparable to or better than existing methods.
• The model demonstrated effectiveness in segmenting polyps from colonoscopy images.

Conclusions:

• The developed CAD system shows significant potential for improving colorectal cancer diagnosis and management.
• Future work includes integrating the model into a medical capsule robot for clinical application.
• This advancement could lead to more accessible and efficient colon cancer screening tools.