EUVIMEDEUVIMED entdecken
Crossref

Automated Detection of Intracranial Hemorrhage on Non-Contrast brain CT by using Deep Learning Model

Noor Fatima, Zaufishan Ahmad, Kainat Manshad, Tania Shahzad, Rubab Fatima, Awais Ali

Journal of Pharma and Biomedics · 30.03.2026 · 4 (1`) · 235-240

DOI: 10.56810/jpbm.004.01.0145 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Pharma and Biomedics, online erschienen am 30.03.2026. Verfasst von Noor Fatima und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Intracranial hemorrhage (ICH) is a critical neurological emergency where rapid and accurate diagnosis directly impacts patient outcomes. Non-contrast CT (NCCT) remains the first-line imaging modality, yet manual interpretation is subject to delay, inter-observer variability, and diagnostic error, particularly in high-volume or resource-limited settings. This study evaluated a VGGNet-based deep learning model with transfer learning for automated detection and subtype classification of ICH on NCCT brain scans using a retrospective dataset of 111 scans from Lahore General Hospital, with ground truth labels established by board-certified radiologists for five subtypes: epidural (EDH), subdural (SDH), subarachnoid (SAH), intraventricular (IVH), and intraparenchymal (IPH). The model achieved an overall accuracy of 83.16% and a macro F1-score of 0.5433. IVH and IPH demonstrated strong performance (AUC > 0.93), while SDH detection failed completely (recall = 0.00) due to severe class imbalance (n = 3 samples), and EDH showed poor recall (0.3333). Marked overfitting was observed, with training accuracy exceeding 90% against validation plateauing at 60–65%. These findings indicate that while 2D transfer learning offers proof-of-concept utility for certain ICH subtypes, it is insufficient for robust multi-class classification, particularly for rare hemorrhages. Overcoming data scarcity through multi-center collaboration, adopting 3D architectures for volumetric context, and implementing imbalance-aware training strategies such as focal loss are essential prerequisites for clinical translation.

Zitieren

APA

Fatima, N., Ahmad, Z., Manshad, K., Shahzad, T., Fatima, R., & Ali, A. (2026). Automated Detection of Intracranial Hemorrhage on Non-Contrast brain CT by using Deep Learning Model. Journal of Pharma and Biomedics, 4(1`), 235-240. https://doi.org/10.56810/jpbm.004.01.0145

Vancouver

Fatima N, Ahmad Z, Manshad K, Shahzad T, Fatima R, Ali A. Automated Detection of Intracranial Hemorrhage on Non-Contrast brain CT by using Deep Learning Model. Journal of Pharma and Biomedics. 2026;4(1`):235-240. doi:10.56810/jpbm.004.01.0145

BibTeX
@article{Fatima2026,
  author = {Fatima, Noor and Ahmad, Zaufishan and Manshad, Kainat and Shahzad, Tania and Fatima, Rubab and Ali, Awais},
  title = {Automated Detection of Intracranial Hemorrhage on Non-Contrast brain CT by using Deep Learning Model},
  journal = {Journal of Pharma and Biomedics},
  year = {2026},
  volume = {4},
  number = {1`},
  pages = {235--240},
  doi = {10.56810/jpbm.004.01.0145}
}
Kostenlos starten →