Deep Bayesian segmentation for colon polyps: Well-calibrated predictions in medical imaging

Daniela L. Ramos,Hector J. Hortua

2024-07-24

Abstract:Colorectal polyps are generally benign alterations that, if not identified promptly and managed successfully, can progress to cancer and cause affectations on the colon mucosa, known as adenocarcinoma. Today advances in Deep Learning have demonstrated the ability to achieve significant performance in image classification and detection in medical diagnosis applications. Nevertheless, these models are prone to overfitting, and making decisions based only on point estimations may provide incorrect predictions. Thus, to obtain a more informed decision, we must consider point estimations along with their reliable uncertainty quantification. In this paper, we built different Bayesian neural network approaches based on the flexibility of posterior distribution to develop semantic segmentation of colorectal polyp images. We found that these models not only provide state-of-the-art performance on the segmentation of this medical dataset but also, yield accurate uncertainty estimates. We applied multiplicative normalized flows(MNF) and reparameterization trick on the UNET, FPN, and LINKNET architectures tested with multiple backbones in deterministic and Bayesian versions. We report that the FPN + EfficientnetB7 architecture with MNF is the most promising option given its IOU of 0.94 and Expected Calibration Error (ECE) of 0.004, combined with its superiority in identifying difficult-to-detect colorectal polyps, which is effective in clinical areas where early detection prevents the development of colon cancer.

Image and Video Processing,Artificial Intelligence,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper mainly aims to address the following issues: ### Research Background and Objectives - **Problem Definition**: In colonoscopy, timely and accurate identification of colon polyps is crucial for preventing colorectal cancer. However, traditional visual diagnostic methods have a certain misdiagnosis rate (6%-28%), which limits the effective detection of early-stage cancer. - **Research Objectives**: To develop a deep learning-based method to improve the accuracy of automatic detection of colon polyps and to quantify prediction uncertainty by introducing Bayesian Neural Networks (BNN) to provide more reliable and well-calibrated prediction results. ### Solution - **Method**: A series of Bayesian Neural Networks based on different convolutional neural network architectures (such as UNET, FPN, LINKNET, etc.) were constructed, combined with various posterior distribution representation techniques (such as Multiplicative Normalizing Flows [MNF] and reparameterization tricks). - **Evaluation Metrics**: Intersection over Union (IOU) was used as the main performance metric, while recall was also considered to comprehensively evaluate the model's performance. Additionally, Expected Calibration Error (ECE) was used to measure the reliability of the model's predictions. ### Main Findings - **Best Model**: Experiments showed that the FPN + EfficientNetB7 architecture combined with MNF technology performed the best on the test set, achieving an IOU value of 0.94 and an ECE value of 0.004. This indicates that the model can not only accurately segment polyp regions but also accurately estimate uncertainty. - **Uncertainty Quantification**: By introducing Bayesian methods, researchers were able to obtain more reliable uncertainty estimates, which is particularly important for clinical decision-making, especially in cases where early detection can significantly improve survival rates. In summary, the goal of this paper is to improve the automatic detection system for colon polyps by utilizing Bayesian Neural Networks, aiming to provide a solution that has both high accuracy and reliable uncertainty estimation.

Deep Bayesian segmentation for colon polyps: Well-calibrated predictions in medical imaging

A Local Region Proposals Approach to Instance Segmentation for Intestinal Polyp Detection

Automatic polyp image segmentation and cancer prediction based on deep learning

ABC-Net: Area-Boundary Constraint Network With Dynamical Feature Selection for Colorectal Polyp Segmentation

Probabilistic Modeling Ensemble Vision Transformer Improves Complex Polyp Segmentation

An Ensemble Framework of Deep Neural Networks for Colorectal Polyp Classification

Dual encoder–decoder-based deep polyp segmentation network for colonoscopy images

Deep learning uncertainty and confidence calibration for the five-class polyp classification from colonoscopy

An Interpretable Deep Neural Network for Colorectal Polyp Diagnosis under Colonoscopy

BetterNet: An Efficient CNN Architecture with Residual Learning and Attention for Precision Polyp Segmentation

MBFFNet: Multi-Branch Feature Fusion Network for Colonoscopy

Uncertainty and interpretability in convolutional neural networks for semantic segmentation of colorectal polyps

Towards a Computed-Aided Diagnosis System in Colonoscopy: Automatic Polyp Segmentation Using Convolution Neural Networks

Focus U-Net: A novel dual attention-gated CNN for polyp segmentation during colonoscopy

Dual-branch multi-information aggregation network with transformer and convolution for polyp segmentation

MCE-Net: polyp segmentation with multiple branch series-parallel attention and channel interaction via edge distribution guidance

Enhanced accuracy with Segmentation of Colorectal Polyp using NanoNetB, and Conditional Random Field Test-Time Augmentation

Bionnica: A Deep Neural Network Architecture for Colorectal Polyps' Premalignancy Risk Evaluation

Simultaneous segmentation and classification of colon cancer polyp images using a dual branch multi-task learning network

Colorectal Polyp Segmentation Using a Fully Convolutional Neural Network

M3ResU-Net: a Deep Residual Network for Multi-Center Colorectal Polyp Segmentation Based on Multi-Scale Learning and Attention Mechanism.