Deep Learning Applied to Intracranial Hemorrhage Detection.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Medicine
• Neurology

Background:

• Intracranial hemorrhage necessitates prompt diagnosis and treatment.
• Accurate identification of hemorrhage location and type is crucial for patient management.
• Manual diagnosis of urgent hemorrhages is time-consuming and challenging for experts.

Purpose of the Study:

• To develop a deep learning-based decision-support system for diagnosing intracranial hemorrhage.
• To improve the efficiency and accuracy of hemorrhage detection in computed tomography (CT) scans.

Main Methods:

• Utilized EfficientDet deep-learning architecture for hemorrhage detection.
• Developed a two-fold methodology: slice-level classification and patient-level hemorrhage evaluation.
• Employed Grad-CAM for visual explanations of classification decisions.

Main Results:

• Achieved 92.7% accuracy in classifying CT scan slices for hemorrhage presence.
• Obtained a Receiver Operating Characteristic Area Under the Curve (ROC AUC) of 0.978 for patient-level hemorrhage diagnosis.
• Provided visual interpretability of the deep learning model's predictions.

Conclusions:

• The proposed EfficientDet-based method shows significant potential as a decision-support system for intracranial hemorrhage diagnosis.
• The AI tool can assist clinicians in rapidly identifying patients requiring urgent procedures.
• Visual explanations enhance the trustworthiness and clinical utility of the AI system.