Source-linked AI summary
SA-Profile: Automated Sulcus Angle Profiling from Super-Resolution MRI
Michael Wehrli, Leo Widmer, Edwin Li, Noel Fiechter, Lorenzo Pettinari, Sidaty El Hadramy, Carol C. Hasler, Philippe C. Cattin
TL;DR
Single-slice sulcus-angle assessment is sensitive to slice selection and landmark placement, limiting characterization of changing trochlear morphology. The paper reconstructs super-resolved MR volumes and automatically fits continuous sulcus-angle profiles, achieving 11.6° mean absolute error against manual measurements and distinct cohort-level profiles. The authors present the framework as exploratory and call for validation in larger, better-annotated cohorts.
Problem
Single-slice sulcus-angle assessment is sensitive to slice selection, landmark placement, and proximal-distal morphological variation.
Method
The framework fuses multi-view MR scans with an implicit neural representation, detects trochlear landmarks with sagittal and axial networks, and fits splines for continuous profiles.
Results
11.6° was the mean absolute error relative to manual measurements, and the fastMRI and in-house TD cohorts showed distinct mean sulcus-angle profiles.
Takeaways & Limitations
Continuous profiling extends conventional single-slice assessment and reveals morphological patterns not captured by one measurement while remaining linked to established sulcus-angle interpretation.
Takeaways & Limitations
The pathological cohort is small, fastMRI is not a healthy reference population, and clinical evaluation used variable single-rater routine measurements.
Abstract
from arXiv · showhide
Trochlear dysplasia (TD) is an abnormality of the femoral trochlea associated with anterior knee pain and patellar instability. The sulcus angle (SA) is used to assess trochlear morphology, but it is typically measured on a single axial MR slice with no clear guidance on which to select, making it sensitive to slice selection and landmark placement. We propose an automatic framework for continuous SA profiling from super-resolved MR volumes. Clinically acquired axial, coronal, and sagittal MR scans are combined using implicit neural representations to reconstruct a high-resolution volume. SA measurements are computed across the trochlear region using two landmark detection U-Net models. The approach was evaluated on the public fastMRI dataset and a small in-house cohort of patients with TD. Compared with conventional manual single-slice SA measurements, the proposed automated method yielded a mean absolute error of 11.6$^\circ$ while providing continuous characterization of trochlear morphology. Population-level analysis demonstrated distinct mean SA profiles between the public cohort and the in-house TD cohort, highlighting the potential of profile-based assessment to characterize TD. By reducing reliance on a single manually selected axial slice, the proposed framework extends conventional SA assessment to a continuous profile-based description of trochlear morphology without additional imaging, while remaining conceptually linked to current clinical assessment. Further validation is required. The code is available: https://github.com/wehrlimi/SA_Profile.
1 Introduction
Trochlear dysplasia assessment commonly relies on a manually selected axial sulcus-angle slice, although slice choice, landmark placement, and proximal-distal morphology limit this scalar measurement. The paper combines super-resolution reconstruction and automated landmark detection to produce a continuous sulcus-angle profile from standard MR acquisitions.
- Trochlear dysplasia is associated with anterior knee pain, patellar instability, and increased dislocation risk, making reproducible morphology assessment important.
- Single-slice sulcus-angle assessment varies with slice selection, observer placement, anatomical reference, and proximal-distal trochlear morphology.Cartilage and subchondral-bone references can differ by 18°.
- Existing automated landmark methods largely preserve predefined-slice dependence, while implicit neural representations enable high-resolution reconstruction from complementary MR acquisitions.
- The proposed framework reconstructs a super-resolved volume, detects sagittal and axial landmarks, and fits anatomy-aligned splines to extract a continuous, interpretable sulcus-angle profile.
2 Methods
The method combines multi-view MR super-resolution, two-stage landmark detection, and spline fitting to generate continuous sulcus-angle profiles across the trochlea. Heatmap distributions are retained during fitting so landmark trajectories remain anatomically continuous across slices.
- The pipeline reconstructs a multi-view super-resolution volume, detects sagittal and axial landmarks, and fits splines for the continuous sulcus-angle profile.
- Super-Resolution: Axial, coronal, and sagittal acquisitions are fused with an implicit neural representation into an isotropic 256^3 volume.The representation is trained with pixel-wise mean-squared reconstruction error.
- Landmark Detection: The sagittal U-Net identifies proximal and distal bounds S1 and S2, while the axial U-Net localizes medial, deepest, and lateral trochlear points T1–T3.
- Landmark Detection: Complete axial heatmaps are retained as spatial keypoint distributions rather than reduced to independent point estimates.
- Spline Fitting: Adaptive spline fitting across T1–T3 enforces anatomical continuity and reduces slice-to-slice variability before sulcus angles are computed.
3 Experiments
Experiments used patient-level separated public and in-house datasets for landmark development, clinical consistency analysis, keypoint evaluation, and cohort-level profile comparison. Manual measurements were obtained at the conventional slice location, while profiles were normalized between sagittal bounds.
- Datasets: The study used 1,566 fastMRI patients with suitable scans and 32 in-house patients with measurable trochlear dysplasia.The fastMRI population is heterogeneous and not a strictly healthy reference cohort.
- Landmark Development: Landmark-model labels covered 159 patients, with 111 training, 16 validation, and 32 patient-level test cases.
- Clinical Evaluation: Clinical consistency was assessed on 25 additional fastMRI and 25 additional in-house patients using manual measurements approximately 3 cm above the joint line.These patients were excluded from landmark-model development and evaluation.
- Evaluation: Landmark performance was measured by Euclidean distance between predicted and ground-truth keypoints in pixels and millimeters.
- Population Analysis: Cohort profiles were normalized from S2 to S1 and summarized by mean and standard deviation along the trochlea.
4 Results
The automated method broadly agreed with conventional manual measurements while revealing patient-specific and cohort-level variation across the trochlea. The in-house TD cohort showed larger mean sulcus angles over a substantial normalized trochlear extent.
- Clinical Consistency: 11.6° was the overall mean absolute error between automated and manual sulcus-angle measurements at corresponding slice locations.Manual means were 150.9° for fastMRI and 164.0° for the in-house TD cohort; automated means were 139.8° and 163.0°, respectively.
- Patient Profiles: Patient-specific profiles included relatively flat, gradually increasing, and proximally flattened patterns among in-house TD patients.
- Keypoint Performance: Sagittal boundary localization errors averaged 4.74 mm for S1 and 2.62 mm for S2, while T1–T3 errors ranged from 2.59 mm to 3.26 mm.
- Population Analysis: The in-house TD cohort had larger mean sulcus-angle values than fastMRI across a substantial part of the normalized trochlear extent.
5 Discussion
The method produces continuous SA profiles with broad agreement to manual measurements, while exposing cohort-level and patient-specific morphology that single-slice assessment cannot capture. Interpretation remains exploratory because reconstruction, cohort composition, and the clinical reference standard impose important limitations.
- 11.6° MAE was observed against manual SA measurements, supporting agreement with current clinical practice rather than comparison with definitive ground truth.Reported observer variability and landmark-placement differences reach up to 9° and 18°, respectively.
- Continuous profiles revealed morphological patterns across the trochlea that a single measurement cannot capture while remaining linked to established SA interpretation.
- Landmark errors of 2.6 mm to 4.7 mm exceeded sub-millimeter single-slice results but still generated profiles broadly consistent with manual measurements.The models operated across the full proximal-distal extent, where appearance and landmark position varied substantially.
- Distinct mean profiles occurred between fastMRI and the in-house TD cohort, particularly across approximately 40–80% of normalized slice position.Because fastMRI is heterogeneous, its profile should be treated as a clinical reference rather than a normative healthy curve.
- The approach remains exploratory because reconstruction may smooth cartilage boundaries, the TD cohort is small, fastMRI is not a healthy reference, and evaluation used variable single-rater measurements.Future validation is planned in larger clinically phenotyped cohorts with multiple expert raters.
6 Conclusion
The study presents an automated framework for continuous sulcus angle profiling from super-resolution knee MRI. It extends single-slice assessment along the trochlear axis, shows broad agreement with manual measurements and additional population-level variation, but requires further validation.
- The framework extracts continuous sulcus angle profiles from super-resolution knee MRI along the proximal-distal trochlear axis.
- The method showed technical feasibility, broad agreement with manual measurements, and population-level morphological variation beyond a single slice.
- Further validation in larger clinically phenotyped cohorts and with multiple expert raters is required.