Real Time Multi Organ Classification on Computed Tomography Images

TL;DR

This work tackles real-time organ labeling in CT by replacing full-volume segmentation with a classifier-based approach using sparse, context-rich descriptors called 3.5D sampling. A 1D residual network classifies pointwise descriptors, and segmentation is achieved by querying grid locations at multiple resolutions, enabling rapid coarse-to-fine masks without heavy resampling. On BTCV data, per-point inference runs around 0.92 ms, with coarse-to-fine segmentation on CPU in seconds, establishing a practical runtime benchmark. The results demonstrate a favorable runtime-accuracy trade-off for real-time clinical workflows and point toward broader applications in detection, registration, and landmarking, with future work extending to more organs and architectures.

Abstract

Organ segmentation is a fundamental task in medical imaging since it is useful for many clinical automation pipelines. However, some tasks do not require full segmentation. Instead, a classifier can identify the selected organ without segmenting the entire volume. In this study, we demonstrate a classifier based method to obtain organ labels in real time by using a large context size with a sparse data sampling strategy. Although our method operates as an independent classifier at query locations, it can generate full segmentations by querying grid locations at any resolution, offering faster performance than segmentation algorithms. We compared our method with existing segmentation techniques, demonstrating its superior runtime potential for practical applications in medical imaging.

Figures (4)

• Figure 1: Descriptor Definition and Decoding
• Figure 2: 1D Residual Network on Sampled Intensities
• Figure 3: Classifier based segmentation enables segmentation in any resolution thus allowing multiple steps of refinement for fine segmentation
• Figure 4: Qualitative Results on BTCV dataset