Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Uncertainty Quantification using Variational Inference for Biomedical Image Segmentation

Abhinav Sagar

TL;DR

This paper addresses uncertainty quantification in biomedical image segmentation by introducing a Bayesian encoder-decoder that learns a posterior over network weights to capture both aleatoric and epistemic uncertainty. The model adopts a variational inference framework inspired by VAEs, using a pre-trained backbone and encoder–decoder architecture to produce a mean and standard deviation for latent variables, enabling sampling and uncertainty estimation. Evaluations on the BRATS brain tumor dataset demonstrate competitive segmentation performance via $DSC$ and $IoU$, while providing uncertainty maps obtained through MC dropout, deep ensembles, and ensemble MC dropout. The work highlights the practical value of principled uncertainty in safety-critical clinical contexts and lays groundwork for uncertainty-aware medical image analysis.

Abstract

Deep learning motivated by convolutional neural networks has been highly successful in a range of medical imaging problems like image classification, image segmentation, image synthesis etc. However for validation and interpretability, not only do we need the predictions made by the model but also how confident it is while making those predictions. This is important in safety critical applications for the people to accept it. In this work, we used an encoder decoder architecture based on variational inference techniques for segmenting brain tumour images. We evaluate our work on the publicly available BRATS dataset using Dice Similarity Coefficient (DSC) and Intersection Over Union (IOU) as the evaluation metrics. Our model is able to segment brain tumours while taking into account both aleatoric uncertainty and epistemic uncertainty in a principled bayesian manner.

Uncertainty Quantification using Variational Inference for Biomedical Image Segmentation

TL;DR

This paper addresses uncertainty quantification in biomedical image segmentation by introducing a Bayesian encoder-decoder that learns a posterior over network weights to capture both aleatoric and epistemic uncertainty. The model adopts a variational inference framework inspired by VAEs, using a pre-trained backbone and encoder–decoder architecture to produce a mean and standard deviation for latent variables, enabling sampling and uncertainty estimation. Evaluations on the BRATS brain tumor dataset demonstrate competitive segmentation performance via and , while providing uncertainty maps obtained through MC dropout, deep ensembles, and ensemble MC dropout. The work highlights the practical value of principled uncertainty in safety-critical clinical contexts and lays groundwork for uncertainty-aware medical image analysis.

Abstract

Deep learning motivated by convolutional neural networks has been highly successful in a range of medical imaging problems like image classification, image segmentation, image synthesis etc. However for validation and interpretability, not only do we need the predictions made by the model but also how confident it is while making those predictions. This is important in safety critical applications for the people to accept it. In this work, we used an encoder decoder architecture based on variational inference techniques for segmenting brain tumour images. We evaluate our work on the publicly available BRATS dataset using Dice Similarity Coefficient (DSC) and Intersection Over Union (IOU) as the evaluation metrics. Our model is able to segment brain tumours while taking into account both aleatoric uncertainty and epistemic uncertainty in a principled bayesian manner.

Paper Structure

This paper contains 15 sections, 22 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Medical Image Segmentation
  4. Bayesian Neural Network
  5. Variational Inference
  6. Aleatoric uncertainty and epistemic uncertainty
  7. Method
  8. Dataset
  9. Data Augmentation
  10. Network Architecture
  11. Experimental Details
  12. Loss Functions
  13. Evaluation Metrics
  14. Results
  15. Conclusions

Figures (4)

  • Figure 1: Example of MRI slices and ground truth segmentation
  • Figure 2: Illustration of our network architecture
  • Figure 3: Accuracy vs data distribution of our network using uncertainty quantification approach on validation/test sets.
  • Figure 4: Examples of models predictions on test samples, compared to ground truth segmentation. First column: input image, second column: ground truth segmentation, third column: predicted segmentation, fourth column: aleatoric uncertainty and fifth column: epistemic uncertainty.