A computer-aided tool for automatic volume estimation of hematoma using non-contrast brain CT scans

Area of Science:

• Medical Imaging
• Radiology
• Computational Pathology

Background:

• Intracerebral hemorrhage (ICH) diagnosis relies on non-contrast computed tomography (NCCT).
• Accurate hematoma volume estimation is critical for ICH treatment planning.
• Current methods may require manual intervention or advanced computational resources.

Purpose of the Study:

• To develop an automatic, computer-aided tool for estimating hematoma volume from 3D CT scans.
• To integrate multiple abstract splitting (MAS) and seeded region growing (SRG) for a unified detection pipeline.
• To validate the proposed method's accuracy and efficiency against existing approaches and deep learning models.

Main Methods:

• A novel pipeline combining MAS and SRG for automatic hematoma detection in pre-processed 3D CT volumes.
• Testing the methodology on 80 ICH patient cases.
• Comparison of estimated volumes against manually segmented ground truth and the ABC/2 method.
• Benchmarking against the U-Net deep learning model for performance evaluation.

Main Results:

• The proposed algorithm achieved an R² correlation coefficient of 0.86 with ground-truth volumes, comparable to the ABC/2 method.
• Unsupervised approach results were similar to supervised U-Net models.
• Average computation time was 132.76 ± 14 seconds.
• The method demonstrated fast and automatic hematoma volume estimation.

Conclusions:

• The developed methodology provides a fast, automatic, and accurate estimation of hematoma volume from 3D CT scans.
• The approach is comparable to established methods and deep learning models but requires less computational power.
• Recommended for clinical practice as a computer-assistive tool for hematoma volume estimation on simple computer systems.