BATseg: Boundary-aware Multiclass Spinal Cord Tumor Segmentation on 3D MRI Scans

TL;DR

BATseg addresses multiclass spinal cord tumor segmentation on 3D MRI by introducing a boundary-aware approach that learns a class-specific tumor surface distance field, guided by a boundary-focused loss in addition to standard segmentation losses. A large-scale dataset of 653 patients with four tumor types is released to enable rigorous evaluation, and BATseg demonstrates superior segmentation performance on both the spinal cord dataset and the KNIGHT kidney tumor dataset. The work highlights the value of explicit boundary modeling for small, morphologically diverse tumors and provides a reproducible baseline with open data and code. This approach has potential to improve morphometric quantification and treatment planning in spinal oncology, while acknowledging limitations in cross-scanner generalizability that warrant further multi-institutional validation.

Abstract

Spinal cord tumors significantly contribute to neurological morbidity and mortality. Precise morphometric quantification, encompassing the size, location, and type of such tumors, holds promise for optimizing treatment planning strategies. Although recent methods have demonstrated excellent performance in medical image segmentation, they primarily focus on discerning shapes with relatively large morphology such as brain tumors, ignoring the challenging problem of identifying spinal cord tumors which tend to have tiny sizes, diverse locations, and shapes. To tackle this hard problem of multiclass spinal cord tumor segmentation, we propose a new method, called BATseg, to learn a tumor surface distance field by applying our new multiclass boundary-aware loss function. To verify the effectiveness of our approach, we also introduce the first and large-scale spinal cord tumor dataset. It comprises gadolinium-enhanced T1-weighted 3D MRI scans from 653 patients and contains the four most common spinal cord tumor types: astrocytomas, ependymomas, hemangioblastomas, and spinal meningiomas. Extensive experiments on our dataset and another public kidney tumor segmentation dataset show that our proposed method achieves superior performance for multiclass tumor segmentation.

Figures (6)

• Figure 1: An illustration of four types of spinal cord tumors shown in the first row, and other commonly studied organs/tumors shown in the second row.
• Figure 2: An illustration of the overall framework.
• Figure 3: The proposed segmentation pipeline. A 3D volume $T$ is fed into the backbone network, predicting per-voxel multiclass results $S$ via the segmentation head, and estimating the tumor surface distance field $F$ via a newly added head.
• Figure 4: An illustration of calculating tumor surface distance values on a single 2D slice.
• Figure 5: The two-stage baseline for our problem of multiclass 3D segmentation. Given an input 3D volume, class-agnostic tumor voxels are segmented against the background in Stage 1, followed by Stage 2 where the tumor type is classified for the same 3D volume. The predicted tumor type is then assigned to the estimated tumor mask.