Robust Quantification of Affected Brain Volume from Computed Tomography Perfusion: A Hybrid Approach Combining Deep...

Area of Science:

• Neuroimaging
• Medical Artificial Intelligence
• Stroke Imaging

Background:

• Accurate volumetric estimation of affected brain tissue in acute ischemic stroke (AIS) is critical for treatment decisions.
• Commercial software for computed tomography perfusion (CTP) analysis has limitations, including variability due to image quality.
• Existing methods may not provide sufficiently robust predictions of infarct core and penumbra.

Purpose of the Study:

• To develop and validate a hybrid deep learning (DL) and machine learning (ML) model for accurate and robust volumetric estimation of affected brain volumes in AIS using CTP.
• To compare the performance of the proposed hybrid model against a leading commercial software solution (RapidAI).

Main Methods:

• A hybrid approach integrating singular value decomposition (SVD), DL (CNN), and ML was developed.
• A CNN model predicted eight vascular landmarks from 449 CTP images of AIS patients.
• SVD-based methods generated perfusion maps, and results were compared with RapidAI using concordance correlation coefficients (CCC) and accuracy metrics.

Main Results:

• The CNN model achieved a mean Euclidean distance error of 4.63 ± 2.00 mm for vessel localization.
• The hybrid model achieved CCC scores of 0.905 for cerebral blood flow (CBF) < 30% and 0.879 for Tmax > 6 s, outperforming the non-ML version.
• The overall data assessment accuracy for the hybrid model reached 0.8.

Conclusions:

• The proposed hybrid DL and ML model offers a robust and accurate method for volumetric estimation in CTP analysis for AIS.
• This approach demonstrates improved performance and robustness compared to current commercial software, potentially enhancing clinical decision-making.