EUVIMEDEUVIMED entdecken
Multizentrische StudieValidierungsstudieOpen Access

Improving the Robustness of Deep Learning Models in Predicting Hematoma Expansion from Admission Head CT

Anh T Tran, Gaby Abou Karam iD, Dorin Zeevi, Adnan I Qureshi iD, Ajay Malhotra iD, Shahram Majidi iD, Santosh B Murthy, Soojin Park, Despina Kontos, Guido J Falcone, Kevin N Sheth iD, Seyedmehdi Payabvash iD

American Journal of Neuroradiology · 2025

DOI: 10.3174/ajnr.a8650 ↗

Kurz gefasst

Multizentrische Studie in der Zeitschrift American Journal of Neuroradiology, 2025. Verfasst von Anh T Tran und 11 weiteren Autor:innen. Beteiligte Einrichtung: From the Department of Radiology (A.T.T., D.Z., S.P.), NewYork-Presbyterian/Columbia University Irving Medical Center, Columbia University, New York,…. Hauptthemen laut MEDLINE: Cerebral Hemorrhage, Deep Learning, Hematoma, Tomography, X-Ray Computed. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 36 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar. Gefördert von: Doris Duke Charitable Foundation, NIH.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

BACKGROUND AND PURPOSE: Robustness against input data perturbations is essential for deploying deep learning models in clinical practice. Adversarial attacks involve subtle, voxel-level manipulations of scans to increase deep learning models' prediction errors. Testing deep learning model performance on examples of adversarial images provides a measure of robustness, and including adversarial images in the training set can improve the model's robustness. In this study, we examined adversarial training and input modifications to improve the robustness of deep learning models in predicting hematoma expansion (HE) from admission head CTs of patients with acute intracerebral hemorrhage (ICH). MATERIALS AND METHODS: We used a multicenter cohort of n = 890 patients for cross-validation/training, and a cohort of n = 684 consecutive patients with ICH from 2 stroke centers for independent validation. Fast gradient sign method (FGSM) and projected gradient descent (PGD) adversarial attacks were applied for training and testing. We developed and tested 4 different models to predict ≥3 mL, ≥6 mL, ≥9 mL, and ≥12 mL HE in an independent validation cohort applying receiver operating characteristics area under the curve (AUC). We examined varying mixtures of adversarial and nonperturbed (clean) scans for training as well as including additional input from the hyperparameter-free Otsu multithreshold segmentation for model. RESULTS: When deep learning models trained solely on clean scans were tested with PGD and FGSM adversarial images, the average HE prediction AUC decreased from 0.8 to 0.67 and 0.71, respectively. Overall, the best performing strategy to improve model robustness was training with 5:3 mix of clean and PGD adversarial scans and addition of Otsu multithreshold segmentation to model input, increasing the average AUC to 0.77 against both PGD and FGSM adversarial attacks. Adversarial training with FGSM improved robustness against similar type attack but offered limited cross-attack robustness against PGD-type images. CONCLUSIONS: Adversarial training and inclusion of threshold-based segmentation as an additional input can improve deep learning model robustness in prediction of HE from admission head CTs in acute ICH.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Cerebral HemorrhageDeep LearningHematomaTomography, X-Ray ComputedAgedFemaleHumansMaleMiddle AgedCohort StudiesPredictive Value of TestsReproducibility of Results

Autor:innen und Institutionen

  • From the Department of Radiology (A.T.T., D.Z., S.P.), NewYork-Presbyterian/Columbia University Irving Medical Center, Columbia University, New York, New York.
  • Department of Radiology and Biomedical Imaging (A.T.T., G.A.K., D.Z., A.M.), Yale School of Medicine, New Haven, Connecticut.
  • Zeenat Qureshi Stroke Institute and Department of Neurology (A.I.Q.), University of Missouri, Columbia, Missouri.
  • Department of Neurosurgery, Icahn School of Medicine at Mount Sinai (S.M.), Mount Sinai Hospital, New York, New York.
  • Department of Neurology (S.B.M.), Weill Cornell Medical College, Cornell University, New York.
  • Department of Neurology (S.P., D.K.), NewYork-Presbyterian/Columbia University Irving Medical Center, Columbia University, New York, New York.

ORCID-Profile: Gaby Abou Karam, Adnan I Qureshi, Ajay Malhotra, Shahram Majidi, Kevin N Sheth, Seyedmehdi Payabvash

Förderung

  • Doris Duke Charitable Foundation
  • NIH

Zitieren

APA

Tran, A. T., Abou Karam, G., Zeevi, D., Qureshi, A. I., Malhotra, A., Majidi, S., Murthy, S. B., Park, S., Kontos, D., Falcone, G. J., Sheth, K. N., & Payabvash, S. (2025). Improving the Robustness of Deep Learning Models in Predicting Hematoma Expansion from Admission Head CT. American Journal of Neuroradiology. https://doi.org/10.3174/ajnr.a8650

Vancouver

Tran AT, Abou Karam G, Zeevi D, Qureshi AI, Malhotra A, Majidi S, et al. Improving the Robustness of Deep Learning Models in Predicting Hematoma Expansion from Admission Head CT. American Journal of Neuroradiology. 2025. doi:10.3174/ajnr.a8650

BibTeX
@article{Tran2025,
  author = {Tran, Anh T and Abou Karam, Gaby and Zeevi, Dorin and Qureshi, Adnan I and Malhotra, Ajay and Majidi, Shahram and Murthy, Santosh B and Park, Soojin and Kontos, Despina and Falcone, Guido J and Sheth, Kevin N and Payabvash, Seyedmehdi},
  title = {Improving the Robustness of Deep Learning Models in Predicting Hematoma Expansion from Admission Head CT},
  journal = {American Journal of Neuroradiology},
  year = {2025},
  doi = {10.3174/ajnr.a8650}
}
Kostenlos starten →