Regional Analysis of Error Patterns from Automatic Cortical Plate Segmentation in Fetal Brain MRI
Poster No:
1629
Submission Type:
Abstract Submission
Authors:
Bowen Qiu1,2, Andrea Gondova1,3, Sungmin You1,3, Milton Candela-Leal1, Hanuman Verma1, Seungyoon Jeong1,3, P. Ellen Grant1,3,4, Kiho Im1,3,5
Institutions:
1Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School, Boston, MA, USA, 2Key Laboratory for Biomedical Engineering of Ministry of Education, Department of Biomedical Engineering, College of Biomedical Engineering & Instrument Science, Zhejiang University, Hangzhou, Zhejiang, China, 3Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School, Boston, MA, USA, 4Department of Radiology, Boston Children’s Hospital, Harvard Medical School, Boston, MA, USA, 5Department of Pediatrics, Harvard Medical School, Boston, MA, USA
First Author:
Bowen Qiu
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Key Laboratory for Biomedical Engineering of Ministry of Education, Department of Biomedical Engineering, College of Biomedical Engineering & Instrument Science, Zhejiang University
Boston, MA, USA|Hangzhou, Zhejiang, China
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Key Laboratory for Biomedical Engineering of Ministry of Education, Department of Biomedical Engineering, College of Biomedical Engineering & Instrument Science, Zhejiang University
Boston, MA, USA|Hangzhou, Zhejiang, China
Co-Author(s):
Andrea Gondova
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School
Boston, MA, USA|Boston, MA, USA
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School
Boston, MA, USA|Boston, MA, USA
Sungmin You
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School
Boston, MA, USA|Boston, MA, USA
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School
Boston, MA, USA|Boston, MA, USA
Milton Candela-Leal
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School
Boston, MA, USA
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School
Boston, MA, USA
Hanuman Verma
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School
Boston, MA, USA
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School
Boston, MA, USA
Seungyoon Jeong
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School
Boston, MA, USA|Boston, MA, USA
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School
Boston, MA, USA|Boston, MA, USA
P. Ellen Grant
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School|Department of Radiology, Boston Children’s Hospital, Harvard Medical School
Boston, MA, USA|Boston, MA, USA|Boston, MA, USA
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School|Department of Radiology, Boston Children’s Hospital, Harvard Medical School
Boston, MA, USA|Boston, MA, USA|Boston, MA, USA
Kiho Im
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School|Department of Pediatrics, Harvard Medical School
Boston, MA, USA|Boston, MA, USA|Boston, MA, USA
Fetal Neonatal Neuroimaging and Developmental Science Center, Harvard Medical School|Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School|Department of Pediatrics, Harvard Medical School
Boston, MA, USA|Boston, MA, USA|Boston, MA, USA
Introduction:
Fetal brain segmentation, with a specific focus on the cortical plate (CP), is essential for the early detection of atypical brain development. Recent advances, deep learning approaches such as convolutional neural networks (CNNs) especially (Dou et al., 2021; Hong et al., 2020; Khalili et al., 2019), have significantly improved quality and speed of automated segmentation. However, current models still show significant inaccuracies in certain regions that impact segmentation reliability. Here, we present a method for analyzing error patterns from CP segmentation using an attention-gated UNet model (Ronneberger et al., 2015; You et al., 2024) and compare these patterns with those derived from a novel fuzzy UNet model (Nan et al., 2022; Price et al., 2019), in order to find out clusters with largest segmentation error.
Methods:
This study, approved by the Institutional Review Board at Boston Children's Hospital, retrospectively analyzed T2-weighted fetal brain MRI from 128 typically developing fetuses (gestational weeks [GW]: 29.36 ± 3.79, range: 21.86 - 37.30). 14 randomly selected cases were used for testing, while the rest were used for training segmentation models. Data was processed using our fetal brain MRI processing pipeline (You et al., 2024), which includes brain masking, non-uniformity correction, slice-to-volume registration, and alignment to the 31-week template. We then analyzed the segmentation error patterns and compared the performance of attention-gated UNet model (You et al., 2024) with a fuzzy UNet model (Nan et al., 2022; Price et al., 2019).
Approach for analysis of segmentation error patterns is summarized in Figure 1. In short, we calculated average False-Positive (FP) and False-Negative (FN) maps across each image pair in the testing set. We identified ground truth edges with morphological operations and generated Distance-Error (DE) maps by calculating the Euclidean distance from each misclassified voxel to the nearest edge. Error map voxel values were then projected to the CP skeleton. We then identified the clusters with highest segmentation errors by filtering the top 50% error values on the skeleton and ranking them based on a summed error values within connected regions. Finally, we compared the within-cluster segmentation quality between attention-gated UNet and fuzzy UNet using the Dice Coefficient, Hausdorff Distance and Hybrid metrics using paired t-tests.
Approach for analysis of segmentation error patterns is summarized in Figure 1. In short, we calculated average False-Positive (FP) and False-Negative (FN) maps across each image pair in the testing set. We identified ground truth edges with morphological operations and generated Distance-Error (DE) maps by calculating the Euclidean distance from each misclassified voxel to the nearest edge. Error map voxel values were then projected to the CP skeleton. We then identified the clusters with highest segmentation errors by filtering the top 50% error values on the skeleton and ranking them based on a summed error values within connected regions. Finally, we compared the within-cluster segmentation quality between attention-gated UNet and fuzzy UNet using the Dice Coefficient, Hausdorff Distance and Hybrid metrics using paired t-tests.
·Figure 1: Pipeline for analysis of segmentation error patterns.
Results:
Identified areas with the largest prevalence of segmentation errors are mainly located around Sylvian Fissure (Figure 2a&b), medial regions of the frontal lobe (figure 2a,b&c), and medial temporal lobe (Figure 2b&c). The average FP and FN errors were positively associated with GW (Linear Mixed-Effect Model, β=2221 and 2031 respectively, p<0.001).
Comparing segmentation results between the attention-gated UNet and the fuzzy UNet, attention-gated UNet performed significantly better in terms of FP and DE only in medial regions of the frontal lobe (ΔHausdorff Distance=1.199, p<0.05; ΔDice=0.026, p<0.05, respectively), while fuzzy UNet outperformed in other metrics such as Hybrid metrics. Overall, these results suggest a comparable segmentation performance between the between the attention-gated UNet and the fuzzy UNet in the error-prone clusters.
Comparing segmentation results between the attention-gated UNet and the fuzzy UNet, attention-gated UNet performed significantly better in terms of FP and DE only in medial regions of the frontal lobe (ΔHausdorff Distance=1.199, p<0.05; ΔDice=0.026, p<0.05, respectively), while fuzzy UNet outperformed in other metrics such as Hybrid metrics. Overall, these results suggest a comparable segmentation performance between the between the attention-gated UNet and the fuzzy UNet in the error-prone clusters.
·Figure 2: Top 5 clusters with the largest FP (a), FN (b), and DE (c) Error and the metric comparisons within those specific clusters.
Conclusions:
We characterized error patterns of automated CP segmentation. As expected, segmentation errors mainly occurred around deep sulci and low-contrast areas. This is probably due to their complex morphology, age-related brain changes, and data quality issues, such as partial volume effects in medial regions. Our segmentation error evaluation provides valuable insights for optimizing and advancing segmentation methods in the future. Building on these findings, we plan to explore advanced segmentation techniques, such as regionally weighted loss, to enhance performance in the identified clusters.
Education, History and Social Aspects of Brain Imaging:
Education, History and Social Aspects of Brain Imaging
Lifespan Development:
Normal Brain Development: Fetus to Adolescence 2
Modeling and Analysis Methods:
Segmentation and Parcellation 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Cortical Anatomy and Brain Mapping
Normal Development
Keywords:
Cortex
Cortical Layers
Machine Learning
MRI
Segmentation
STRUCTURAL MRI
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Hong, J. (2020). Fetal Cortical Plate Segmentation Using Fully Convolutional Networks With Multiple Plane Aggregation. Frontiers in Neuroscience, 14, 591683. https://doi.org/10.3389/fnins.2020.591683
Khalili, N. (2019). Automatic brain tissue segmentation in fetal MRI using convolutional neural networks. Artificial Intelligence in MRI, 64, 77–89. https://doi.org/10.1016/j.mri.2019.05.020
Nan, Y. (2022). Fuzzy Attention Neural Network to Tackle Discontinuity in Airway Segmentation (arXiv:2209.02048). arXiv. http://arxiv.org/abs/2209.02048
Price, S. R. (2019). Introducing Fuzzy Layers for Deep Learning. 2019 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), 1–6. https://doi.org/10.1109/FUZZ-IEEE.2019.8858790
Ronneberger, O. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation (arXiv:1505.04597). arXiv. https://doi.org/10.48550/arXiv.1505.04597
You, S. (2024). Automatic cortical surface parcellation in the fetal brain using attention-gated spherical U-net. Frontiers in Neuroscience, 18, 1410936. https://doi.org/10.3389/fnins.2024.1410936