Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-0592
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-07-30T06:09:04.945615Z
status: published
lens: g_space
sector: health
headline: Deep learning models for predicting clinical activity and characteristics in thyroid eye disease using magnetic resonance imaging
dek: Purpose This project aims to develop and evaluate deep learning models using orbital magnetic resonance imaging for the prediction of continuous clinical activity score and key patient characteristics in thyroid eye disease. Methods The publicly available TOM500 dataset, consisting of orbital magnetic resonance imaging scans and clinical data from 500 thyroid eye disease patients, was split into training ( n = 360), validation ( n = 100), and test ( n = 40) sets. A ResNet-50 convolutional neural network pretrain…
gain_title: Deep learning models using orbital MRI predicted continuous clinical activity score with low error and inferred patient characteristics including age, sex, and smoking status.
problem_title: (none)
trace_subject: (none)
gain_reading: Deep learning models using orbital MRI predicted continuous clinical activity score with low error and inferred patient characteristics including age, sex, and smoking status.
gain_evidence: The clinical activity score prediction model achieved a mean absolute error of 1.05 | The age model achieved a mean absolute error of 6.90 years, while the sex and smoking status models achieved an accuracy of 93% and 80%, respectively
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers developed and evaluated ResNet-50 models on the TOM500 orbital MRI dataset to predict continuous clinical activity score and patient characteristics in thyroid eye disease, training on 360 cases, validating on 100, and testing on 40.

The reported test performance suggests MRI contains learnable signals for disease activity and demographics, but the small test set and single public dataset leave generalizability, clinical utility, and comparison to standard clinical assessment unproven.
limitation: 
tag: Evidence-backed gain
key_points: Study used publicly available TOM500 dataset of 500 thyroid eye disease patients split into training n=360, validation n=100, and test n=40 | Adapted ResNet-50 pretrained on ImageNet for four tasks with each slice converted into 3-channel input of raw scan, segmentation mask, and weighted average | Sex and smoking status models reported AUC of 0.974 and 0.747 respectively
rundown: The authors used the TOM500 dataset of 500 patients and adapted a ResNet-50 CNN pretrained on ImageNet, converting each MRI slice into a 3-channel input comprising the raw scan, segmentation mask, and weighted average.

Performance was reported on a 40-patient test set with MAE of 1.05 for clinical activity score, MAE of 6.90 years for age, 93% accuracy and 0.974 AUC for sex, and 80% accuracy and 0.747 AUC for smoking status.
sources:
- peer_reviewed | Orbit | https://doi.org/10.1080/01676830.2026.2708890 | 2026-07-28
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

1e9b129c08bbd6b06f50b5fec481c8422bf450e87e14a3952ba217026d069436

previous

0000000000000000000000000000000000000000000000000000000000000000