Title: Enhancing radiology workflow: alert system for intracranial hemorrhages using deep learning and single-board computers

Authors: Shanu Nizarudeen; Ganesh Ramaswamy Shanmughavel

Addresses: Department of Electronics, Noorul Islam Centre for Higher Education, Kumaracoil, Tamil Nadu, India ' Department of Electronics, R.M.K. Engineering College, Chennai, Tamil Nadu, India

Abstract: This study investigates the application of low-cost embedded platforms integrating deep learning models for intracranial hemorrhage classification using head CT images Two experiments were conducted: transfer learning feature extraction and benchmarking CTNet against four AI architectures TL achieved 96% accuracy in multi-label classification with ROC-AUC scores of 0.99-0.997 EfficientNetB0 outperformed other methods in classifying hemorrhage subtypes CTNet showed excellent grading performance in most bleeding types, but relatively weaker performance in subarachnoid hemorrhage. Models were implemented on a Raspberry Pi using TensorFlow Lite for real-time prediction and audio notifications image acquisition conditions' impact on results was addressed CTNet demonstrated continuous improvement during training This research highlights the potential of low-cost embedded platforms with deep learning models for optimising workflow in emergency clinics, providing faster interpretation and improving patient care.

Keywords: intracranial hemorrhage; ICH; single-board computers; classification; transfer learning; deep learning; embedded system.

DOI: 10.1504/IJSCC.2024.138534

International Journal of Systems, Control and Communications, 2024 Vol.15 No.2, pp.122 - 145

Received: 11 Jul 2023
Accepted: 14 Sep 2023
Published online: 10 May 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article