AI-Enhanced Interface for Colonic Polyp Segmentation Using DeepLabv3+ with Comparative Backbone Analysis | JoVE Visualize

Area of Science:

Medical Imaging
Artificial Intelligence
Oncology

Background:

Colonic polyps are precursors to colorectal cancer, necessitating accurate detection and segmentation for timely intervention.
Manual polyp segmentation in colonoscopy images is time-consuming, prone to human error, and requires expert interpretation.
Automated segmentation methods are crucial for enhancing the speed, accuracy, and reliability of polyp detection.

Purpose of the Study:

To develop and evaluate an automated deep learning model for accurate segmentation of colonic polyps.
To assess the performance of the proposed DeepLabv3+ model with a ResNet-50 backbone using the Kvasir-SEG dataset.
To create a user-friendly Graphical User Interface (GUI) for practical application of the polyp segmentation tool.

Main Methods:

Implementation of the DeepLabv3+ architecture with an encoder-decoder structure and ResNet-50 as the backbone.
Preprocessing of colonoscopy images from the Kvasir-SEG dataset for model training.
Training and rigorous testing of the developed deep learning model, followed by performance metric calculation.

Main Results:

The ResNet-50 based DeepLabv3+ model achieved high segmentation accuracy with a Dice Similarity Coefficient (DSC) of 0.9609 and Mean Intersection over Union (mIoU) of 0.9246.
The model demonstrated significant effectiveness in segmenting colonic polyps, indicating its potential for clinical application.
A functional GUI was developed, facilitating the practical use of the automated segmentation tool for colonoscopy images.

Conclusions:

The proposed DeepLabv3+ model with a ResNet-50 backbone provides a highly accurate and effective solution for colonic polyp segmentation.
Automated polyp segmentation holds substantial promise for improving the early diagnosis of colorectal cancer and optimizing surgical planning for polypectomy.
The developed tool can significantly aid clinicians by providing rapid and reliable image analysis, potentially reducing diagnostic errors and improving patient outcomes.