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

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.

Ähnliche Arbeiten