A revolutionary acute subdural hematoma detection based on two-tiered artificial intelligence model

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Acute subdural hemorrhages require rapid diagnosis for effective treatment.
• Current diagnostic methods can be time-consuming, delaying critical interventions.
• There is a need for advanced tools to assist physicians in interpreting medical images.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) based hybrid method for the detection of acute subdural hemorrhages.
• To assess the AI model's performance in interpreting tomography images for hemorrhage identification.
• To provide a tool for rapid physician warning in suspected cases.

Main Methods:

• A two-level hybrid AI model was developed, combining deep learning (Mask R-CNN) for hemorrhage segmentation and machine learning (Support Vector Machines - SVM) for classification.
• The Mask R-CNN model generated masks for hemorrhagic regions.
• Optimized SVM algorithms, tuned using the bee colony algorithm, performed binary classification.

Main Results:

• The Mask R-CNN model achieved a mean average precision (mAP) of 0.754 at an IOU of 0.5.
• A 5-fold cross-validation yielded an mAP of 0.736.
• The two-level classification process demonstrated 96.36% accuracy, with low false-negative (6.20%) and false-positive (2.57%) rates.

Conclusions:

• The proposed AI model successfully detects hemorrhages and highlights suspicious areas in 2D images.
• This method offers a low-cost, high-accuracy alternative to current telemedicine interpretations.
• The AI system can significantly aid physicians in the timely diagnosis of acute subdural hemorrhages.