Development of Diagnostic Support Software for Phase Estimation of Intracerebral Hemorrhage Using Multi-sequence...

Intracerebral hemorrhage is a severe type of stroke with high morbidity and mortality. Accurate assessment of hemorrhage phase is essential for determining treatment strategies, but magnetic resonance imaging evaluation remains subjective and lacks standardized criteria. This study aimed to create an objective method for estimating the phase of intracerebral hemorrhage using multi-sequence magnetic resonance imaging with 6 sequences: T1-weighted imaging, T2-weighted imaging, diffusion-weighted imaging, apparent diffusion coefficient map, fluid-attenuated inversion recovery, and T2 star-weighted magnetic resonance angiography. We retrospectively analyzed 56 patients with intracerebral hemorrhage. Magnetic resonance images used in this study were acquired using GE 1.5T scanners. Relative signal intensities were calculated using the pons as a reference, and a hemorrhage map was generated using color-coding pixels according to the 5 hemorrhage phases. Seven observers performed visual evaluations under 2 conditions: conventional image sets and hemorrhage maps. Diagnostic performance was assessed using precision-recall curves, average precision, and mean average precision. Interobserver agreement was evaluated using Fleiss' κ coefficient. The mean average precision of visual evaluation with the hemorrhage map was significantly higher than that of visual evaluation with original images (0.81 vs. 0.57, p < 0.01). Visual evaluation with hemorrhage map achieved "Almost perfect agreement" (κ = 0.85), whereas visual evaluation with original images demonstrated only "Slight agreement" (κ = 0.06). This study developed a novel diagnostic support method for estimating intracerebral hemorrhage phase using multi-sequence magnetic resonance imaging. By visualizing relative signal intensity as a color-coded hemorrhage map, the proposed method significantly improved both diagnostic accuracy and interobserver agreement compared with conventional visual evaluation.