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

Journal: Scientific reports
PMID:

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.

Authors

  • Wi-Sun Ryu
    Artificial Intelligence Research Center, JLK Inc., 5 Teheran-ro 33-gil, Seoul, Republic of Korea. wisunryu@gmail.com.
  • Dawid Schellingerhout
    Departments of Cancer Systems Imaging and Neuroradiology, MD Anderson Cancer Center, The University of Texas, Houston, Texas, USA.
  • Jonghyeok Park
    Artificial Intelligence Research Center (W.-S.R, J.P.), JLK Inc, Seoul, Republic of Korea.
  • Jinyong Chung
    National Priority Research Center for Stroke and Department of Neurology (W.-S.R, J.C., S.-W.J., D.-S.G., D.-E.K.), Dongguk University Ilsan Hospital, Goyang, Republic of Korea.
  • Sang-Wuk Jeong
    National Priority Research Center for Stroke and Department of Neurology (W.-S.R, J.C., S.-W.J., D.-S.G., D.-E.K.), Dongguk University Ilsan Hospital, Goyang, Republic of Korea.
  • Dong-Seok Gwak
    National Priority Research Center for Stroke and Department of Neurology (W.-S.R, J.C., S.-W.J., D.-S.G., D.-E.K.), Dongguk University Ilsan Hospital, Goyang, Republic of Korea.
  • Beom Joon Kim
    Departments of Dermatology, Chung-Ang University College of Medicine, Seoul, Korea.
  • Joon-Tae Kim
    Department of Neurology (J.-T.K., M.S.P.,), Chonnam National University Hospital, Gwangju, Republic of Korea.
  • Keun-Sik Hong
    Department of Neurology (K.-S.H., Y,-J.C.), Inje University Ilsan Paik Hospital, Goyang, Republic of Korea.
  • Kyung Bok Lee
    Department of Neurology (K.B.L.), Soonchunhyang University Hospital, Seoul, Republic of Korea.
  • Tai Hwan Park
    Department of Neurology (T.H.P.), Seoul Medical Center, Seoul, Republic of Korea.
  • Sang-Soon Park
    Department of Neurology, Seoul Medical Center, Seoul, South Korea.
  • Jong-Moo Park
    Department of Neurology (J.-M.P.), Uijeongbu Eulji Medical Center, Uijeongbu, Republic of Korea.
  • Kyusik Kang
    Department of Neurology (K.K.), Nowon Eulji Medical Center, Eulji University School of Medicine, Seoul, Republic of Korea.
  • Yong-Jin Cho
    Department of Neurology (K.-S.H., Y,-J.C.), Inje University Ilsan Paik Hospital, Goyang, Republic of Korea.
  • Hong-Kyun Park
    Department of Neurology, Inje University Ilsan Paik Hospital, Inje University College of Medicine, Goyang, South Korea.
  • Byung-Chul Lee
    Genoplan Korea Inc., Seoul, Republic of Korea. io@genoplan.com.
  • Kyung-Ho Yu
    Department of Neurology (B.-C.L., K.-H.Y., M.S.O.), Hallym University Sacred Heart Hospital, Anyang, Republic of Korea.
  • Mi Sun Oh
    Department of Neurology (B.-C.L., K.-H.Y., M.S.O.), Hallym University Sacred Heart Hospital, Anyang, Republic of Korea.
  • Soo Joo Lee
    Department of Neurology (S.J.L.), Eulji University Hospital, Daejeon, Republic of Korea.
  • Jae Guk Kim
    Department of Emergency Medicine, College of Medicine, Hallym University, Chuncheon, South Korea.
  • Jae-Kwan Cha
    Department of Neurology (J.-K.C., D.-H.K.), Dong-A University Hospital, Busan, Republic of Korea.
  • Dae-Hyun Kim
    Department of Neurology (J.-K.C., D.-H.K.), Dong-A University Hospital, Busan, Republic of Korea.
  • Jun Lee
    Nephrology Unit, Department of Medicine, Sarawak General Hospital, Sarawak, Malaysia.
  • Man Seok Park
    Department of Neurology (J.-T.K., M.S.P.,), Chonnam National University Hospital, Gwangju, Republic of Korea.
  • Dongmin Kim
    JLK, Incorporated, Eonju-ro, Gangnam-gu, Seoul, South Korea.
  • Oh Young Bang
    Department of Neurology Samsung Medical Center Sungkyunkwan University School of Medicine Seoul Republic of Korea.
  • Eung Yeop Kim
    Department of Radiology, Gil Medical Center, Gachon University College of Medicine, Incheon, South Korea.
  • Chul-Ho Sohn
    Department of Radiology, Seoul National University Hospital, Seoul, Republic of Korea; Department of Radiology, Seoul National University College of Medicine, Seoul, Republic of Korea; Institute of Radiation Medicine, Seoul National University Medical Research Center, Seoul, Republic of Korea.
  • Hosung Kim
    Laboratory of Neuro Imaging, Keck School of Medicine of USC, University of Southern California, Los Angeles, CA, USA. Electronic address: hosung.kim@loni.usc.edu.
  • Hee-Joon Bae
  • Dong-Eog Kim
    National Priority Research Center for Stroke and Department of Neurology (W.-S.R, J.C., S.-W.J., D.-S.G., D.-E.K.), Dongguk University Ilsan Hospital, Goyang, Republic of Korea kdongeog@duih.org.