KEVS: Enhancing Segmentation of Visceral Adipose Tissue in Pre-Cystectomy CT with Gaussian Kernel Density Estimation

TL;DR

KEVS is designed to predict visceral adipose tissue (VAT) in pre-cystectomy CT scans without requiring ground-truth VAT masks. It combines a semantic segmentation network (U-Mamba nnU-Net) trained on open datasets (SAROS and TotalSegmentator) to identify abdominal cavity and SAT, with Gaussian Kernel Density Estimation (GKDE) applied to SAT intensities to refine VAT predictions within the organ-free cavity; a 15% GKDE percentile is used to remove unlikely voxels. On a dataset of 20 pre-cystectomy scans (UCLH-Cyst) with expert VAT masks, KEVS achieves state-of-the-art Dice Coefficient (~0.87–0.87) and related metrics, outperforming HU-threshold methods and TotalSegmentator. The method reduces inter-observer variability, adapts to scan dose, and is trained entirely on open-source data, with potential to improve peri-operative risk assessment and broader CT-based body tissue analysis.

Abstract

Purpose: The distribution of visceral adipose tissue (VAT) in cystectomy patients is indicative of the incidence of post-operative complications. Existing VAT segmentation methods for computed tomography (CT) employing intensity thresholding have limitations relating to inter-observer variability. Moreover, the difficulty in creating ground-truth masks limits the development of deep learning (DL) models for this task. This paper introduces a novel method for VAT prediction in pre-cystectomy CT, which is fully automated and does not require ground-truth VAT masks for training, overcoming aforementioned limitations. Methods: We introduce the Kernel density Enhanced VAT Segmentator ( KEVS), combining a DL semantic segmentation model, for multi-body feature prediction, with Gaussian kernel density estimation analysis of predicted subcutaneous adipose tissue to achieve accurate scan-specific predictions of VAT in the abdominal cavity. Uniquely for a DL pipeline, KEVS does not require ground-truth VAT masks. Results: We verify the ability of KEVS to accurately segment abdominal organs in unseen CT data and compare KEVS VAT segmentation predictions to existing state-of-the-art (SOTA) approaches in a dataset of 20 pre-cystectomy CT scans, collected from University College London Hospital (UCLH-Cyst), with expert ground-truth annotations. KEVS presents a 4.80% and 6.02% improvement in Dice Coefficient over the second best DL and thresholding-based VAT segmentation techniques respectively when evaluated on UCLH-Cyst. Conclusion: This research introduces KEVS; an automated, SOTA method for the prediction of VAT in pre-cystectomy CT which eliminates inter-observer variability and is trained entirely on open-source CT datasets which do not contain ground-truth VAT masks.

Figures (5)

• Figure 1: A depiction of the full implementation of KEVS. First, a U-Mamba architecture trained on our extended TotalSegmentator dataset makes a prediction on a CT scan. The abdominal cavity and L3 axial SAT prediction are extracted simultaneously, the SAT prediction is eroded and a GKDE fit to the resulting pixel intensities. This GKDE is used to predict the voxels in the abdominal cavity mask which are least likely to be VAT, which are removed from the abdominal cavity mask in the final step to give an informed prediction for VAT.
• Figure 2: (a) An example of the TotalSegmentator dataset for a coronal, axial, and sagittal view of a thoracic/abdominal CT scan. (b) An example of the combined SAROS predictions and TotalSegmentator ground-truth.
• Figure 3: (a) An axial CT slice at the predicted L3 level with predicted SAT (purple) mask. (b) The same CT with the SAT mask eroded to $20\%$ of its original area. (c) A visualisation of the GKDE fit to the eroded SAT mask.
• Figure 4: (a) Prediction of VAT inside abdominal organs for (-190, -30) thresholding.(b) An example of false-negative VAT prediction (yellow) around abdominal organs for TotalSegmentator model. In both cases, these erroneous predictions are reasons for reduction in predictive performance when compared to KEVS.
• Figure 5: (a) A comparison of ground-truth mask, KEVS, and TotalSegmentator prediction on an axial slice (b) A comparison of ground-truth mask, KEVS, and (-190,-30) thresholding prediction on an axial slice. In both cases, the white boxes outline examples of KEVS better performance in comparison to the alternative method.