Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DINOv3 with Test-Time Calibration for Automated Carotid Intima-Media Thickness Measurement on CUBS v1

Zhenpeng Zhang, Jinwei Lu, Yurui Dong, Bo Yuan

Abstract

Carotid intima-media thickness (CIMT) measured from B-mode ultrasound is an established vascular biomarker for atherosclerosis and cardiovascular risk stratification. Although a wide range of computerized methods have been proposed for carotid boundary delineation and CIMT estimation, robust and transferable deep models that jointly address segmentation and measurement remain underexplored, particularly in the era of vision foundation models. Motivated by recent advances in adapting DINOv3 to medical segmentation and exploiting DINOv3 in test-time optimization pipelines, we investigate a DINOv3-based framework for carotid intima-media complex segmentation and subsequent CIMT measurement on the Carotid Ultrasound Boundary Study (CUBS) v1 dataset. Our pipeline predicts the intima-media band at a fixed image resolution, extracts upper and lower boundaries column-wise, corrects for image resizing using the per-image calibration factor provided by CUBS, and reports CIMT in physical units. Across three patient-level test splits, our method achieved a mean test Dice of 0.7739 $\pm$ 0.0037 and IoU of 0.6384 $\pm$ 0.0044. The mean CIMT absolute error was 181.16 $\pm$ 11.57 $μ$m, with a mean Pearson correlation of 0.480 $\pm$ 0.259. In a held-out validation subset ($n=28$), test-time threshold calibration reduced the mean absolute CIMT error from 141.0 $μ$m at the default threshold to 101.1 $μ$m at the measurement-optimized threshold, while simultaneously reducing systematic bias toward zero. Relative to the error ranges reported in the original CUBS benchmark for classical computerized methods, these results place a DINOv3-based approach within the clinically relevant $\sim$0.1 mm measurement regime. Together, our findings support the feasibility of using vision foundation models for interpretable, calibration-aware CIMT measurement.

DINOv3 with Test-Time Calibration for Automated Carotid Intima-Media Thickness Measurement on CUBS v1

Abstract

Carotid intima-media thickness (CIMT) measured from B-mode ultrasound is an established vascular biomarker for atherosclerosis and cardiovascular risk stratification. Although a wide range of computerized methods have been proposed for carotid boundary delineation and CIMT estimation, robust and transferable deep models that jointly address segmentation and measurement remain underexplored, particularly in the era of vision foundation models. Motivated by recent advances in adapting DINOv3 to medical segmentation and exploiting DINOv3 in test-time optimization pipelines, we investigate a DINOv3-based framework for carotid intima-media complex segmentation and subsequent CIMT measurement on the Carotid Ultrasound Boundary Study (CUBS) v1 dataset. Our pipeline predicts the intima-media band at a fixed image resolution, extracts upper and lower boundaries column-wise, corrects for image resizing using the per-image calibration factor provided by CUBS, and reports CIMT in physical units. Across three patient-level test splits, our method achieved a mean test Dice of 0.7739 0.0037 and IoU of 0.6384 0.0044. The mean CIMT absolute error was 181.16 11.57 m, with a mean Pearson correlation of 0.480 0.259. In a held-out validation subset (), test-time threshold calibration reduced the mean absolute CIMT error from 141.0 m at the default threshold to 101.1 m at the measurement-optimized threshold, while simultaneously reducing systematic bias toward zero. Relative to the error ranges reported in the original CUBS benchmark for classical computerized methods, these results place a DINOv3-based approach within the clinically relevant 0.1 mm measurement regime. Together, our findings support the feasibility of using vision foundation models for interpretable, calibration-aware CIMT measurement.
Paper Structure (20 sections, 2 equations, 3 figures, 2 tables)

This paper contains 20 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Carotid ultrasound segmentation and CIMT measurement
  4. Deep learning for carotid wall analysis
  5. Vision foundation models and DINOv3 in medical imaging
  6. Materials and Methods
  7. Dataset
  8. Patient-level split protocol
  9. DINOv3 segmentation model
  10. CIMT extraction and resize-aware unit correction
  11. Test-time calibration
  12. Evaluation metrics
  13. Results
  14. Segmentation performance
  15. CIMT measurement in physical units
  16. ...and 5 more sections

Figures (3)

  • Figure 1: End-to-end CIMT measurement workflow. The segmented intima-media band is converted into upper and lower boundaries, transformed into a column-wise thickness profile, and finally converted from resized-pixel units to physical units using calibration-factor correction.
  • Figure 2: Quantitative performance summary of the proposed DINOv3-based CIMT measurement system. The figure highlights absolute measurement error, systematic bias, agreement behavior, and the improvement obtained by test-time calibration.
  • Figure 3: Additional qualitative evidence for DINOv3-based CIMT measurement. The top panel summarizes the effect of threshold calibration on representative cases, while the lower panels provide complementary visual evidence of model response patterns and challenging residual-error cases.