Video Summarization and Fracture Detection in Pediatric Wrist Ultrasound Using Deep Reinforcement Learning
Bibliographic record
Abstract
Wrist fractures are common injuries among children, significantly impacting daily activities and contributing to prolonged wait times in emergency departments. Portable ultrasound is a safe, radiation-free, and cost-effective diagnostic tool with real-time imaging capabilities to identify these fractures. Ultrasound (US) videos offer dynamic, detailed views for fracture detection but can be lengthy and redundant, making interpretation time-consuming for clinicians. To address this challenge, we developed a novel video summarization and fracture detection method using Deep Reinforcement Learning (DRL). The DRL agent emulates a human expert by analyzing the entire video, selecting diagnostically relevant frames, and leveraging a CNN to classify these frames as either normal or fractured. The reward mechanism, which prioritizes feature similarity and frame dissimilarity, enhances the agent's ability to identify the keyframes. Anisotropic diffusion is applied to US images to enhance bright bony regions, before feature extraction and representativeness calculation. Frame similarity calculations are parallelized to reduce computational complexity, storage demands and simplify the classification task by focusing solely on frames indicative of fractures. On a dataset of 114 patients, the proposed classification network achieved an accuracy of 88.8% with a sensitivity of 92.5% and a specificity of 86.6% using the RL-generated video summaries. Our primary focus is on enhancing sensitivity to ensure that no fractured cases are missed. This performance surpasses the 84.4% accuracy achieved with full video classification. AI-driven ultrasound, with its efficiency, and reduced computational demands, provides a promising solution for early disease detection in resource-constrained environments.Clinical relevance- This AI-driven fast, cost-effective ultrasound tool achieves high accuracy and sensitivity while reducing training time by 50%, making it suitable for real-time clinical use on low-power devices. This can assist lightly-trained healthcare providers in early disease detection, reduce long wait times, and enhance treatment access in remote areas.
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 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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".