Evaluation of Traumatic Subdural Hematoma Volume by Using Image Segmentation Assessment Based on Deep Learning

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Medicine
• Neurosurgery

Background:

• Accurate traumatic subdural hematoma (TSDH) volume measurement is crucial for effective patient treatment.
• Current methods for TSDH volume assessment may lack speed or precision.
• Deep learning offers potential for automated and accurate medical image analysis.

Purpose of the Study:

• To evaluate the consistency of TSDH volume measurements using a deep learning-based convolutional neural network (CNN) algorithm.
• To compare the performance of CNN segmentation against manual segmentation and the ABC/2 formula.
• To assess the efficiency and accuracy of CNN for TSDH volume calculation.

Main Methods:

• A retrospective study analyzed 90 CT images from patients diagnosed with TSDH.
• Hematoma volumes were measured using manual segmentation, CNN algorithm segmentation, and the ABC/2 formula.
• Consistency was tested against manual segmentation as the gold standard; statistical analysis included percentage error and area under the curve (AUC).

Main Results:

• The CNN algorithm demonstrated a lower percentage error compared to the ABC/2 formula.
• No significant difference was found between CNN segmentation and manual segmentation (P > 0.05).
• AUC values indicated strong performance for all methods, with manual segmentation (0.840) and CNN (0.832) closely comparable and outperforming ABC/2 (0.811).

Conclusions:

• The CNN-based algorithm provides efficient and accurate segmentation for TSDH volume calculation.
• This deep learning approach shows high consistency with manual segmentation, validating its clinical utility.
• CNN offers a reliable tool for rapid and precise TSDH volume assessment in clinical practice.