RADnet: Radiologist level accuracy using deep learning for hemorrhage detection in CT scans
importance 6/10
partially solved
Brain Tumor Detection and Classification medical imaging stage: research
Problem
Automatically detecting brain hemorrhage in CT scans to support radiologist diagnosis.
Why it matters
Fast, accurate hemorrhage detection is critical for emergency diagnosis and patient outcomes.
What the AI did
RADnet, a deep learning model, emulated the radiologist workflow by scanning 2D slices with attention and using recurrent 3D context to produce CT-level hemorrhage predictions.
Results
81.82% hemorrhage prediction accuracy on 77 brain CTs, benchmarked against independent analysis by three senior radiologists.
Impact
Peer-reviewed with independent radiologist benchmarking; clinical adoption not stated in source.
Record
| Venue | IEEE International Symposium on Biomedical Imaging (ISBI) |
|---|---|
| Published | 2018 |
| Work type | conference-paper |
| Authors | Monika Grewal, Muktabh Mayank Srivastava, Pulkit Kumar, Srikrishna Varadarajan |
| Citations | 213 (OpenAlex, at import time) |
| References | 17 |
| Model or method | RADnet (Recurrent Attention DenseNet) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1109/isbi.2018.8363574 |
| Open access | green |
| Keywords | deep learning, CT scans, hemorrhage detection, DenseNet, medical imaging |
Importance rating
6
significant — thousands to millions affected
Abstract gives no scale or impact evidence; importance inferred from supporting radiologist CT diagnosis of hemorrhage.