Sep 09, 2024
Deep-learning tool for early identification of non-traumatic intracranial hemorrhage etiology based on CT scan
- 1Beijing Tiantan Hospital
Protocol Citation: Meng Zhao 2024. Deep-learning tool for early identification of non-traumatic intracranial hemorrhage etiology based on CT scan. protocols.io https://dx.doi.org/10.17504/protocols.io.yxmvmejoog3p/v1
License: This is an open access protocol distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Protocol status: Working
We use this protocol and it's working
Created: September 08, 2024
Last Modified: September 09, 2024
Protocol Integer ID: 107136
Abstract
Objective
To develop an
artificial intelligence system that can accurately identify acute non-traumatic
intracranial hemorrhage (ICH) etiology based on non-contrast CT (NCCT) scans
and investigate whether clinicians can benefit from it in a diagnostic setting.
Methods
The deep learning model
was developed with 1868 eligible NCCT scans with non-traumatic ICH collected
between January 2011 and April 2018. We tested the model on two independent
datasets (TT200 and SD 98) collected after April 2018. The model’s diagnostic
performance was compared with clinicians’ performance. We further designed a
simulated study to compare the clinicians’ performance with and without the
deep learning system augmentation.
Results
The proposed deep
learning system achieved area under the receiver operating curve of
0.986 (95% CI 0.967–1.000) on aneurysms, 0.952 (0.917–0.987) on hypertensive
hemorrhage, 0.950 (0.860–1.000) on arteriovenous malformation (AVM), 0.749
(0.586–0.912) on Moyamoya disease (MMD), 0.837 (0.704–0.969) on cavernous
malformation (CM), and 0.839 (0.722–0.959) on other causes in TT200 dataset.
Given a 90% specificity level, the sensitivities of our model were 97.1% and
90.9% for aneurysm and AVM diagnosis, respectively. The model also shows an
impressive generalizability in an independent dataset SD98. The clinicians
achieve significant improvements in the sensitivity, specificity, and accuracy
of diagnoses of certain hemorrhage etiologies with proposed system
augmentation.
Conclusions
The proposed deep learning algorithms can be
an effective tool for early identification of hemorrhage etiologies based on
NCCT scans. It may also provide more information for
clinicians for triage and further imaging examination selection.
Dataset and Clinical Taxonomy Retrospectively review NCCT scans from 4019 patients at Beijing Tiantan Hospital (2011-2018). Include only the first NCCT scan post-ICH event; exclude traumatic, post-surgical, or delayed scans.
Inclusion and Exclusion Criteria Exclude scans with poor quality, lacking a complete series, or from patients under four years. Review medical records for patient demographics and clinical history.
Independent Test Datasets Collection Collect additional datasets from Beijing Tiantan Hospital (TT200) and Shandong Jining Medical College (SD98) from 2018 to 2019.
Ethical Compliance Obtain IRB approval from both institutions. Anonymize all clinical data and images. Waive informed consent based on non-harm justification.
Data Labeling and Etiologic Classification Use Meretoja et al.'s method for etiologic classification. Label data independently by two radiologists, with discrepancies resolved by a third.
Model Development: Image Preprocessing Rotate images every 20 degrees using a bilinear algorithm. Resize volumes to a consistent voxel size. Apply unsupervised skull-stripping and crop the images.
Model Architecture Develop the ICHNet using the "SlowFast Networks" architecture. Use separate pathways for different resolution features, combining them before the output layer.
Training Protocol Use 4 NVIDIA Tesla P40 GPUs and Pytorch-0.4.1. Implement min-max normalization and oversampling to address class imbalance. Perform five-fold cross-validation to select the best model.
Test Procedure Test the model using an ensemble strategy with data from both test datasets. Decrease rotation frequency during testing to reduce processing time.
Statistical Analysis Generate confusion matrices and calculate AUCs. Use the Clopper-Pearson and Hanley and McNeil methods for confidence intervals. Evaluate rater concordance and assess model impact using Cohen and Fleiss κ coefficients. Apply bootstrapping for performance metrics and Wilcoxon rank-sum tests for statistical significance.