Using a Convolutional Neural Network and Convolutional Long Short-term Memory to Automatically Detect Aneurysms on 2D...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurosurgery

Background:

• 2D digital subtraction angiography (DSA) images lack spatial information, making it difficult to distinguish cerebral aneurysms from overlapping vessels.
• Accurate detection of posterior communicating artery aneurysms is crucial for timely intervention and patient outcomes.

Purpose of the Study:

• To develop a deep learning diagnostic system for enhanced detection of posterior communicating artery aneurysms on 2D DSA images.
• To validate the efficiency and performance of the developed deep learning system in aneurysm detection.

Main Methods:

• A two-stage deep learning system was proposed, including a region localization stage and an intracranial aneurysm detection stage.
• A bi-input+RetinaNet+convolutional long short-term memory (C-LSTM) framework was developed and compared against three existing frameworks using 5-fold cross-validation.
• Performance was assessed using metrics such as AUC, mean average precision, sensitivity, specificity, and accuracy.

Main Results:

• The bi-input+RetinaNet+C-LSTM framework achieved the highest Area Under the Curve (AUC) value of 0.97.
• This framework demonstrated a mean sensitivity of 89% and a mean specificity of 93%, outperforming other models and human experts in certain metrics.
• The system achieved a mean accuracy of 91%, indicating its high efficacy in detecting aneurysms.

Conclusions:

• Incorporating both spatial and temporal information significantly improves the performance of deep learning frameworks for aneurysm detection.
• The proposed bi-input+RetinaNet+C-LSTM system shows superior performance and has the potential to assist physicians in the accurate detection of intracranial aneurysms on 2D DSA images.