Automated Computer-Aided Detection and Classification of Intracranial Hemorrhage Using Ensemble Deep Learning...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Intracranial Hemorrhage (ICH) diagnosis is critical for patient outcomes but remains challenging.
• Early detection of ICH is vital to prevent mortality and morbidity.

Purpose of the Study:

• To develop and evaluate an ensemble deep learning model for accurate and early detection of various types of ICH.
• To improve the classification accuracy of epidural, intraventricular, subarachnoid, intra-parenchymal, and subdural hemorrhages.

Main Methods:

• Utilized an ensemble of Convolutional Neural Networks (CNNs), specifically Squeeze and Excitation-based Residual Networks (SE-ResNeXT) and Long Short-Term Memory (LSTM) networks.
• Employed windowing for preprocessing and data augmentation on the RSNA and CQ500 datasets.
• Implemented Gradient-weighted Class Activation Mapping (Grad-CAM) for region of interest identification.

Main Results:

• The proposed ensemble model achieved an overall accuracy of 99.79% and an F-score of 0.97.
• Individual hemorrhage type classification accuracies exceeded 98%, with some reaching 99.89%.
• Demonstrated superior performance compared to existing deep learning models in ICH detection.

Conclusions:

• The ensemble deep learning approach using SE-ResNeXT and LSTM effectively detects and classifies multiple types of Intracranial Hemorrhage.
• The model's high accuracy and AUC scores indicate its potential for clinical application in diagnosing brain bleeds.
• This method offers a promising advancement in automated medical image analysis for neurological emergencies.