Hip osteoporosis risk from CT imaging and clinical data
Lead researcher, under Dr. Champa Jayasuriya
A prior result did not survive honest re-evaluation, and a simpler model proved more trustworthy.
88%approx., identifying at-risk patients
The problem
Osteoporosis usually goes undiagnosed until a fracture happens. A hip CT taken for another reason already contains bone-density information; the question is whether it can be combined with routine clinical variables to flag risk earlier.
Screening normally means a DEXA scan — a separate appointment a patient has to be referred to and turn up for. Reading a CT they have already had removes both steps: the scan exists, and nothing about how it was taken has to change (UToledo News, 8 September 2026).
The data
109 hip CT patients — 71 with osteoporosis, 38 without — from University of Toledo Medical Center, with 15 clinical variables including age, BMI, calcium, haemoglobin and glucose.
The imaging is real patient data and never leaves the institution. Nothing derived from it is published here beyond segmentation masks and schematics.
A confound found during the audit: some images had been exported at different scan settings, which can bias any texture-based feature.
Approach
U-Net segmentation to isolate bone regions, reaching a Dice coefficient of 0.976, then radiomic feature extraction with PyRadiomics — texture, shape and intensity — fused with the clinical variables.
Six models trained under Bayesian hyperparameter optimisation: logistic regression, SVM, random forest, AdaBoost, CatBoost and a DNN.
As the university describes the tool: bone tissue is isolated from the scan, 18 bone-density features are extracted, and those are combined with age, BMI and lab markers to return a high, medium or low risk rating in minutes (UToledo News, 8 September 2026).
The pipeline was inherited from a prior student and audited end to end before any new result was produced.
Results
The audit found a data-leakage bug: the model was effectively seeing test data during training.
After correction and re-validation, a previously reported headline result did not hold. A simpler, more robust model emerged as the most trustworthy of the six.
A single train/test split was replaced with repeated cross-validation, which showed the initial results were noise-driven and confirmed genuine signal only under stricter testing.
The tool identifies at-risk patients with approximately 88% accuracy, and the next phase targets up to 90%. Both figures are the university’s, published when it wrote the work up (UToledo News, 8 September 2026). This site withheld them until then and quoted no number in their place.
“Approximately 88%” is the article’s wording and is reproduced as such. It is one reported figure with no confidence interval, no per-class breakdown and no held-out cohort stated, so it is not comparable to the cross-validated distributions above and is not presented as one.
Limitations and what I'd do differently
109 patients is a small cohort with an imbalanced split, 71 to 38. Repeated cross-validation gives a distribution rather than a point estimate, but it cannot manufacture statistical power that the sample size does not contain.
The scan-setting confound is flagged, not solved. Stratifying by acquisition parameters, or harmonising features across them, is the correct next step and has not been done.
Single-centre data. Nothing here demonstrates that the result transfers to another hospital’s scanners or population, which is why standardising performance across scanners from different health systems is one of the two things the next phase is for — the other being extending past the upper femur to the hip and spine (UToledo News, 8 September 2026).
The lesson I would carry forward: audit the inherited pipeline before extending it. The leakage was upstream of everything I was asked to add.
Artifacts
- A $50,000 National Science Foundation grant through UToledo’s I-Corps program, funding testing and customer discovery.
- A patent application filed through the University of Toledo Technology Transfer Office — filed, not granted. The granted patent elsewhere on this site is IN 573184 and is unrelated work.
- A manuscript under review. Not a publication until it is accepted, and not counted as one.
- Press: UToledo News, 8 September 2026 — the source for every figure here that is not my own measurement.
Patient imaging is never published in any form.