Deep learning-based automatic segmentation of cerebral infarcts on diffusion MRI

We explored effects of (1) training with various sample sizes of multi-site vs. single-site training data, (2) cross-site domain adaptation, and (3) data sources and features on the performance of algorithms segmenting cerebral infarcts on Magnetic Resonance Imaging (MRI). We used 10,820 annotated diffusion-weighted images (DWIs) from 10 university hospitals. Algorithms based on 3D U-net were trained using progressively larger subsamples (ranging from 217 to 8661), while internal testing employed a distinct set of 2159 DWIs. External validation was conducted using three unrelated datasets (n = 2777, 50, and 250). For domain adaptation, we utilized 50 to 1000 subsamples from the 2777-image external target dataset. As the size of the multi-site training data increased from 217 to 1732, the Dice similarity coefficient (DSC) and average Hausdorff distance (AHD) improved from 0.58 to 0.65 and from 16.1 to 3.75 mm, respectively. Further increases in sample size to 4330 and 8661 led to marginal gains in DSC (to 0.68 and 0.70, respectively) and in AHD (to 2.92 and 1.73). Similar outcomes were observed in external testing. Notably, performance was relatively poor for segmenting brainstem or hyperacute (< 3 h) infarcts. Domain adaptation, even with a small subsample (n = 50) of external data, conditioned the algorithm trained with 217 images to perform comparably to an algorithm trained with 8661 images. In conclusion, the use of multi-site data (approximately 2000 DWIs) and domain adaptation significantly enhances the performance and generalizability of deep learning algorithms for infarct segmentation.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2025-04-16
DOI
https://doi.org/10.1038/s41598-025-91032-w
Citations
9
Primary Topic
Advanced Neuroimaging Techniques and Applications
Type
article
Field-Weighted Citation Impact
12.24

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Deep learning-based automatic segmentation of cerebral infarcts on diffusion MRI

Eung Yeop Kim, Hosung Kim, Mi Sun Oh, Wi‐Sun Ryu et al.
9 citations
Scientific Reports
Advanced Neuroimaging Techniques and Applications
12.24
article

Deep learning-based automatic segmentation of cerebral infarcts on diffusion MRI

Eung Yeop Kim, Hosung Kim, Mi Sun Oh, Wi‐Sun Ryu, Tai Hwan Park, Keun‐Sik Hong, Man‐Seok Park, Soo Joo Lee, Oh Young Bang, Byung‐Chul Lee, Sang‐Soon Park, Jinyong Chung, Dong‐Eog Kim, Jae‐Kwan Cha, Hee‐Joon Bae, Jong-Hyeok Park, Hong‐Kyun Park, Yong‐Jin Cho, Sang‐Wuk Jeong, Jong‐Moo Park, Chul‐Ho Sohn, Dawid Schellingerhout, Kyusik Kang, Dong-Seok Gwak, Dong‐Min Kim, Jun Lee, Beom Joon Kim, Joon‐Tae Kim, Jae Guk Kim, Kyung Bok Lee, Dae-Hyun Kim, Kyung-Ho Yu
article en
9 citations

Abstract

We explored effects of (1) training with various sample sizes of multi-site vs. single-site training data, (2) cross-site domain adaptation, and (3) data sources and features on the performance of algorithms segmenting cerebral infarcts on Magnetic Resonance Imaging (MRI). We used 10,820 annotated diffusion-weighted images (DWIs) from 10 university hospitals. Algorithms based on 3D U-net were trained using progressively larger subsamples (ranging from 217 to 8661), while internal testing employed a distinct set of 2159 DWIs. External validation was conducted using three unrelated datasets (n = 2777, 50, and 250). For domain adaptation, we utilized 50 to 1000 subsamples from the 2777-image external target dataset. As the size of the multi-site training data increased from 217 to 1732, the Dice similarity coefficient (DSC) and average Hausdorff distance (AHD) improved from 0.58 to 0.65 and from 16.1 to 3.75 mm, respectively. Further increases in sample size to 4330 and 8661 led to marginal gains in DSC (to 0.68 and 0.70, respectively) and in AHD (to 2.92 and 1.73). Similar outcomes were observed in external testing. Notably, performance was relatively poor for segmenting brainstem or hyperacute (< 3 h) infarcts. Domain adaptation, even with a small subsample (n = 50) of external data, conditioned the algorithm trained with 217 images to perform comparably to an algorithm trained with 8661 images. In conclusion, the use of multi-site data (approximately 2000 DWIs) and domain adaptation significantly enhances the performance and generalizability of deep learning algorithms for infarct segmentation.

Scientific ReportsVol. 15(1)
University of Southern California (US), The University of Texas MD Anderson Cancer Center (US), Seoul National University (KR), Eulji University (KR), Dong-A University Hospital (KR), Soonchunhyang University Hospital Seoul (KR), Samsung Medical Center (KR), Seoul National University Bundang Hospital (KR), Seoul Medical Center (KR), Sacred Heart Hospital (US), Chonnam National University Hospital (KR), Dongguk University Ilsan Hospital (KR), Inje University Ilsan Paik Hospital (KR), Daejeon Eulji Medical Center, Eulji University (KR), Singapore Bioimaging Consortium (SG), Yeungnam University Medical Center (KR), Hallym University Sacred Heart Hospital (KR), Sungkyunkwan University (KR)
Korea Medical Device Development Fund, National Research Foundation, National Research Foundation of Korea
Openalex Percentile: Top 1%
Advanced Neuroimaging Techniques and Applications
12.24
DOIOpen Access PDFOpenAlex
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.