Image quality assessment for machine learning tasks using meta-reinforcement learning
Problem
Medical images vary in usefulness for downstream ML tasks; a method was needed to assess task amenability and adapt quickly to new datasets.
Why it matters
Unusable images can undermine clinical diagnosis and interventions such as prostate procedures and pneumonia detection.
What the AI did
A neural-network IQA controller trained with the task predictor, plus a meta-reinforcement learning framework enabling efficient fine-tuning on new datasets or meta-tasks.
Results
not stated in source
Impact
Efficacy demonstrated on two clinical applications (ultrasound prostate intervention, X-ray pneumonia detection) per abstract, without numeric results.
Record
| Venue | Medical Image Analysis |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Shaheer U. Saeed, Yunguan Fu, Vasilis Stavrinides, Zachary M. C. Baum, Qianye Yang, Mirabela Rusu, Richard E. Fan, Geoffrey A. Sonn et al. |
| Citations | 50 (OpenAlex, at import time) |
| References | 87 |
| Model or method | Meta-reinforcement learning based neural-network IQA controller |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1016/j.media.2022.102427 |
| Open access | hybrid |
| Keywords | image quality assessment, meta-reinforcement learning, task amenability, ultrasound, X-ray |
Importance rating
real but narrower advance
Task-amenability IQA plus meta-RL adaptation is a methodological advance for medical imaging pipelines; demonstrated on two clinical applications but a narrower enabling-method contribution.