Ultrasound for Automated Classification of Full-Thickness Rotator Cuff Tendon Tears using Deep Learning
Bibliographic record
Abstract
Rotator cuff tendon tears, the most common shoulder injuries, are typically diagnosed mainly through MRI, but can also be seen on ultrasound (US), a much less costly test that currently requires highly-trained human expert operators. An AI tool to identify full-thickness rotator cuff tears in US could make this test much more accessible in clinical practice. We propose a two-step approach starting with segmentation and followed by classification. Automatic segmentation of US scans is challenging due to speckle noise and low contrast. We utilized a CNN-autoencoder that predicts boundary contour points of humeral cortex and subacromial bursa directly from raw US images rather than the popular pixel-wise semantic segmentation. Then both the original US image and the corresponding segmentation mask are passed to a classification network (VGG-16) to determine whether tendons are torn or intact. This novel approach only passes the key portions of the scan (in which any tears are most visible) to the classification network, maximizing detection accuracy and clinical relevance. We evaluated this approach on data prospectively acquired from 210 patients, training with 11,600 images and testing with 2900 images. We had an average segmentation Dice coefficient (DC) of 95.3% and Hausdorff Distance (HD) of 2.9 mm, outperforming a U-Net model (DC=90.5%, HD=6.8 mm). The classification network, VGG-16, achieved 85.2% accuracy (sensitivity 84.2%, specificity 83.3%) in classifying supraspinatus tendons as intact or torn from US images. Results indicate that our AI-driven US evaluation pipeline has the potential to enable less-experienced ultrasound users to detect rotator cuff tears with high accuracy and explainability. This can allow more healthcare professionals to conduct scans, improving timely patient access to imaging and streamlining treatment decisions.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".