Deep Learning Model Based on Transcranial Sonography for Detection of Acute Intracranial Hemorrhage in Adults: A Comparative Study With Sonologists of Different Experience Levels

Background To evaluate the diagnostic performance of a deep learning model for acute intracranial hemorrhage (ICH) on transcranial sonography (TCS) versus sonologists with different levels of experience. Methods We retrospectively studied 395 adults with suspected acute ICH undergoing handheld TCS, using noncontrast head CT as the reference standard (194 ICH). Participants were randomized 8:2 to training (n = 317) and test (n = 78) sets. An EfficientNet-B2 model was trained on cine-loop frames, with frame probabilities averaged per participant. Six blinded sonologists (2 each with > 10, 5–10, and < 5 years’ experience) read the images. Accuracy, sensitivity, specificity, positive and negative predictive values (PPV, NPV), and area under the receiver operating characteristic curve (AUC) were calculated, with calibration and decision curve analyses. Results In the test set (38 ICH, 40 non-ICH), model AUC was 0.956 (95% confidence interval, 0.892–0.996), accuracy 91.03%, sensitivity 89.47%, specificity 92.50%, PPV 91.89%, and NPV 90.24%. For senior, mid-level, and junior sonologists, respectively: accuracy 91.67%/78.85%/66.03%, sensitivity 94.74%/82.89%/69.74%, specificity 88.75%/75.00%/62.50%, PPV 88.89%/75.90%/63.86%, and NPV 94.67%/82.19%/68.49%. Model accuracy matched senior sonologists ( P > 0.05) and exceeded mid-level and junior sonologists ( P = 0.008, P < 0.001); sensitivity and specificity matched senior and mid-level sonologists (all P > 0.05) and exceeded junior sonologists ( P = 0.007, P = 0.001). Calibration was good, and decision curve analysis demonstrated a positive net benefit. Conclusions The TCS-based model showed good discrimination of acute ICH, matching senior and outperforming junior sonologists, and may aid screening in prehospital, bedside, and resource-limited settings.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11612041/v1
Primary Topic
Intracerebral and Subarachnoid Hemorrhage Research
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Deep Learning Model Based on Transcranial Sonography for Detection of Acute Intracranial Hemorrhage in Adults: A Comparative Study With Sonologists of Different Experience Levels

Lianhua Zhu, Xiaomeng Zou, Chiyuan Ma, Xingyao Cui et al.
6 links
Research Square
Intracerebral and Subarachnoid Hemorrhage Research
preprint

Deep Learning Model Based on Transcranial Sonography for Detection of Acute Intracranial Hemorrhage in Adults: A Comparative Study With Sonologists of Different Experience Levels

Lianhua Zhu, Xiaomeng Zou, Chiyuan Ma, Xingyao Cui, Yueming Li, Ruilan Niu, Naiqin Fu, Jianbo Gao, Xu Li, Xiazhen Qin, Zhili Wang, Yongqi Fang
preprint en

Abstract

Background To evaluate the diagnostic performance of a deep learning model for acute intracranial hemorrhage (ICH) on transcranial sonography (TCS) versus sonologists with different levels of experience. Methods We retrospectively studied 395 adults with suspected acute ICH undergoing handheld TCS, using noncontrast head CT as the reference standard (194 ICH). Participants were randomized 8:2 to training (n = 317) and test (n = 78) sets. An EfficientNet-B2 model was trained on cine-loop frames, with frame probabilities averaged per participant. Six blinded sonologists (2 each with > 10, 5–10, and < 5 years’ experience) read the images. Accuracy, sensitivity, specificity, positive and negative predictive values (PPV, NPV), and area under the receiver operating characteristic curve (AUC) were calculated, with calibration and decision curve analyses. Results In the test set (38 ICH, 40 non-ICH), model AUC was 0.956 (95% confidence interval, 0.892–0.996), accuracy 91.03%, sensitivity 89.47%, specificity 92.50%, PPV 91.89%, and NPV 90.24%. For senior, mid-level, and junior sonologists, respectively: accuracy 91.67%/78.85%/66.03%, sensitivity 94.74%/82.89%/69.74%, specificity 88.75%/75.00%/62.50%, PPV 88.89%/75.90%/63.86%, and NPV 94.67%/82.19%/68.49%. Model accuracy matched senior sonologists ( P > 0.05) and exceeded mid-level and junior sonologists ( P = 0.008, P < 0.001); sensitivity and specificity matched senior and mid-level sonologists (all P > 0.05) and exceeded junior sonologists ( P = 0.007, P = 0.001). Calibration was good, and decision curve analysis demonstrated a positive net benefit. Conclusions The TCS-based model showed good discrimination of acute ICH, matching senior and outperforming junior sonologists, and may aid screening in prehospital, bedside, and resource-limited settings.

Research Square
Chinese PLA General Hospital (CN), Chinese General Hospital College of Nursing and Liberal Arts (PH), Chinese People's Liberation Army (CN)
Good health and well-being
Intracerebral and Subarachnoid Hemorrhage Research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon