Diffusion Models for Implicit Image Segmentation Ensembles

TL;DR

This paper addresses semantic segmentation of brain tumors in MRIs with reliable uncertainty estimation. It introduces a diffusion-model–based segmentation framework that conditions a DDPM on the input MRI to produce image-specific masks and a distribution of segmentations. The key contributions are training the DDPM on ground-truth masks with MRI priors, enabling implicit ensemble generation and pixel-wise uncertainty maps, and demonstrating competitive performance on BRATS2020. The work offers a practical approach to uncertainty-aware segmentation in clinical settings, highlighting the trade-off between accuracy and sampling speed.

Abstract

Diffusion models have shown impressive performance for generative modelling of images. In this paper, we present a novel semantic segmentation method based on diffusion models. By modifying the training and sampling scheme, we show that diffusion models can perform lesion segmentation of medical images. To generate an image specific segmentation, we train the model on the ground truth segmentation, and use the image as a prior during training and in every step during the sampling process. With the given stochastic sampling process, we can generate a distribution of segmentation masks. This property allows us to compute pixel-wise uncertainty maps of the segmentation, and allows an implicit ensemble of segmentations that increases the segmentation performance. We evaluate our method on the BRATS2020 dataset for brain tumor segmentation. Compared to state-of-the-art segmentation models, our approach yields good segmentation results and, additionally, detailed uncertainty maps.

Figures (7)

• Figure 1: Workflow for the implicit generation of segmentation ensembles and uncertainty maps with diffusion models. The input image consists of four different MR sequences. Going $n$ times through the sampling process of the diffusion model with different Gaussian noise, $n$ different segmentation masks are generated.
• Figure 2: The training and sampling procedure of our method. In every step $t$, the anatomical information is induced by concatenating the brain MR images $b$ to the noisy segmentation mask $x_{b,t}$.
• Figure 3: Examples of the produced mean and variance maps for 100 sampling runs.
• Figure 4: Performance of the ensemble with respect to the number of samples for the examples $b_1$, $b_2$, and $b_3$, presented in Figure \ref{['varmap']}.
• Figure 5: Comparison of the different uncertainty maps for the three examples.