Detecting Intracranial Hemorrhage with Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Automated detection of intracranial hemorrhage (ICH) is crucial for computer-aided diagnosis systems.
• Previous methods relied on multi-step processes with handcrafted features.
• A deep learning approach offers a more integrated solution.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) for automated ICH detection from CT images.
• To improve upon traditional multi-step detection methods by using an end-to-end learning approach.
• To enhance diagnostic accuracy and efficiency for radiologists.

Main Methods:

• Utilized a deep convolutional neural network (CNN) for simultaneous feature learning and classification.
• Employed data augmentation through image rotations to improve performance.
• Applied postprocessing techniques to the CNN output to increase specificity.

Main Results:

• The CNN achieved 81% sensitivity per lesion and 98% specificity per case on a test set of 134 CT cases.
• Performance demonstrated comparable sensitivity to previous methods but significantly higher specificity.
• The deep learning approach eliminated the need for multiple hand-tuned steps.

Conclusions:

• The developed CNN shows significant promise for accurate and efficient automated detection of intracranial hemorrhage.
• This deep learning model offers a substantial improvement in specificity compared to prior approaches.
• Further performance enhancements are anticipated with database expansion.