A deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans
Problem
Diagnosing acute intracranial hemorrhage on non-contrast head CT varies greatly among radiologists due to subtle findings and heavy time pressure.
Why it matters
Acute ICH is life-threatening and requires rapid, accurate emergency diagnosis.
What the AI did
A deep learning approach combining a 2D CNN with two sequence models detects acute ICH and classifies its five subtypes, trained on the 2019 RSNA Brain CT Hemorrhage Challenge dataset of over 25,000 CT scans.
Results
AUCs of 0.988 (ICH), 0.984 (EDH), 0.992 (IPH), 0.996 (further subtype truncated in abstract)
Impact
High AUCs on a large public challenge dataset demonstrate strong detection performance; clinical deployment not stated.
Record
| Venue | NeuroImage Clinical |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Xiyue Wang, Tao Shen, Sen Yang, Jun Lan, Yanming Xu, Minghui Wang, Jing Zhang, Xiao Han |
| Citations | 192 (OpenAlex, at import time) |
| References | 58 |
| Model or method | 2D CNN plus two sequence models (names not stated) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1016/j.nicl.2021.102785 |
| Open access | gold |
| Keywords | intracranial hemorrhage, head CT, deep learning, CNN, RSNA challenge, AUC |
Importance rating
significant — thousands to millions affected
Acute ICH is described as life-threatening and requiring emergency attention; expert-level detection on a large challenge dataset with external validation.