Spectral adapters improve liver tumor segmentation from CT scans
Spectral Adapters for Segment Anything Model-based Segmentation of Colorectal Liver Metastases in Computed Tomography
Computer Vision and Pattern Recognition
Summary
Segmenting cancer spots in liver CT scans helps doctors plan treatments and track progress. The paper introduces two lightweight add-ons to a powerful image model that improve how well it finds liver tumors. These add-ons make the model better at this task while only needing to adjust a small number of settings. Tests on hundreds of scans show these methods perform close to fully trained traditional models but with fewer training demands.
What this means in practice
- •For medical imaging teams: Enhance colorectal liver metastases segmentation in CT scans with efficient spectral adapters that require fewer trainable parameters.
- •For software developers in medical ai: Integrate spectral adapter techniques to adapt general segmentation models for specialized organ tumor detection tasks.
Authors
Ramtin Mojtahedi, Mohammad Hamghalam, Jacob J. Peoples, Natalie Gangai, Mithat Gonen, Yun Shin Chun, HyunSeon Christine Kang, Richard K. G. Do, Amber L. Simpson
Abstract
Accurate segmentation of colorectal liver metastases (CRLM) in contrast-enhanced computed tomography (CT) is important for response assessment, surgical planning, and follow-up. We propose two parameter-efficient spectral adapters for the Segment Anything Model (SAM): the Directional Spectral Adapter (DiSECT) and Spectral Instance-Guided Adapter (SiGA). DiSECT uses singular value decomposition of frozen weights to constrain residual updates to leading spectral directions, while SiGA adds global and input-conditioned gating through a multilayer perceptron. We evaluate these methods on 446 contrast-enhanced CT volumes (355 training, 91 testing) and compare them with LoRA, QLoRA, convolutional adapters (CAD), and a 3D nnU-Net baseline. Experiments consider single-point, three-point, bounding-box, and no-prompt regimes. SiGA achieves the best single-point performance with a Dice score of 0.77, IoU of 0.69, and HD95 of 35.39 mm. Under no-prompt inference, SiGA reaches 0.76 Dice, 0.68 IoU, and 46.76 mm HD95, comparable to the nnU-Net baseline (0.758 Dice). DiSECT uses only 0.14 million trainable parameters. These results show that spectral adapters can efficiently adapt SAM for CRLM segmentation while retaining strong accuracy with limited trainable parameters.