Automated polyp segmentation for colonoscopy images: A method based on convolutional neural networks and ensemble...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Polyp segmentation in colonoscopy images is crucial for early detection and diagnosis.
• Existing methods may lack efficiency and accuracy in complex polyp segmentation tasks.

Purpose of the Study:

• To develop an efficient and accurate automatic polyp segmentation method for colonoscopy images.
• To leverage ensemble learning with pretrained convolutional neural networks for improved segmentation performance.

Main Methods:

• An ensemble model combining Unet-VGG, SegNet-VGG, and PSPNet was proposed.
• Transfer learning with VGG16 was utilized to create specialized encoder-decoder architectures.
• A weight voting method was employed to ensemble the segmentation results from individual models.

Main Results:

• The proposed method achieved high segmentation accuracy on cvc300, CVC-ClinicDB, and ETIS-LaribPolypDB datasets.
• IoU scores reached up to 96.95% and DICE scores up to 98.45% on the ETIS-LaribPolypDB dataset.
• The ensemble method demonstrated superior performance compared to single-model approaches, with IoU and DICE improvements ranging from 1.98% to 6.38%.

Conclusions:

• The ensemble learning approach successfully enhances automatic polyp segmentation accuracy.
• This method holds potential for improving polyp dataset establishment and clinical applications.
• The proposed technique offers a robust solution for precise polyp lesion area segmentation.