Efficient fine-tuning methods improve liver tumor mapping from CT scans
Parameter-Efficient Fine-Tuning of Foundation Models for Liver Tumor Segmentation in CT
Computer Vision and Pattern Recognition
Summary
Liver tumors need to be accurately found in CT images to help doctors plan treatments. The authors studied ways to adjust a large AI model called SAM to do this better without needing to change all its settings. They tested several techniques that only update small parts of the model, making the process faster and using less computer memory. Their results show that certain methods balance accuracy and efficiency well, which is helpful when computing power or labeled data is limited.
What this means in practice
- •For medical imaging teams: Enhance liver tumor segmentation tools using efficient fine-tuning techniques that reduce adaptation costs on CT scan data.
- •For cloud ai service providers: Offer quicker and lower-resource model customization for medical image analysis services by integrating parameter-efficient fine-tuning strategies.
Authors
Ramtin Mojtahedi, Mohammad Hamghalam, Jacob J. Peoples, Richard K. G. Do, Amber L. Simpson
Abstract
We evaluated parameter-efficient fine-tuning (PEFT) of the Segment Anything Model (SAM) for liver tumor segmentation in abdominal CT of colorectal liver metastases. We compared Low-Rank Adaptation (LoRA), 4-bit Quantized LoRA (QLoRA), a convolutional adapter (Conv-Adapter), and our Directional Spectral Top-K adapter (DiSCo), training only adapters while freezing the SAM backbone. DiSCo derives spectral bases from singular value decomposition of row-normalized weights and learns rank-gated spectral coefficients, per-output magnitude offsets, and a spectral gain, with optional Top-K rank selection at inference and 0.14 M trainable parameters. We benchmarked five prompting regimes: no prompt, single-point, multi-point, and bounding boxes at intersection over union 0.50 and 0.75. Conv-Adapter and LoRA achieved the highest accuracy (overall Dice 0.793 and 0.792; single-point Dice 0.795 and 0.792; 95th-percentile Hausdorff distance (HD95) 32 mm). QLoRA was close (overall Dice 0.766; single-point Dice 0.768; HD95 36.41 mm), with 0.91 M trainable parameters, 120 ms latency, and 4.9 GB peak memory. DiSCo achieved the highest Dice per million trainable parameters (4.66), with overall Dice 0.653, single-point Dice 0.698, and HD95 49.53 mm. These results show an accuracy-efficiency trade-off and support PEFT for liver tumor segmentation with reduced adaptation costs when compute and labeled data are limited. Code: https://github.com/Ramtin-Mojtahedi/PEFT-SAM-Liver-CT