BrainLossNet: a fast, accurate and robust method to estimate brain volume loss from longitudinal MRI

• DOI: 10.1007/s11548-024-03201-3
• Publication Date: 2024-06-16T09:01:44Z
• Authors: Roland Opfer; Julia Krüger; Thomas Buddenkotte; Lothar Spies; Finn Behrendt; Sven Schippling; Ralph Buchert
• Keywords: Interquartile range; Robustness

Abstract Purpose MRI-derived brain volume loss (BVL) is widely used as neurodegeneration marker. SIENA state-of-the-art for BVL measurement, but limited by long computation time. Here we propose “BrainLossNet”, a convolutional neural network (CNN)-based method BVL-estimation. Methods BrainLossNet uses CNN-based non-linear registration of baseline(BL)/follow-up(FU) 3D-T1w-MRI pairs. computed parenchyma masks segmented in the BL/FU scans. The estimate corrected image distortions using apparent change total intracranial volume. was trained on 1525 pairs from 83 scanners. Agreement between and assessed 225 94 MS patients acquired with single scanner 268 52 scanners various indications. Robustness to short-term variability compared 354 healthy men same session without repositioning 116 (Frequently-Traveling-Human-Phantom dataset, FTHP). Results Processing time 2–3 min. median [interquartile range] SIENA-BrainLossNet difference 0.10% [− 0.18%, 0.35%] 0.08% 0.14%, 0.28%] indications dataset. distribution FTHP dataset narrower ( p = 0.036; 95th percentile: 0.20% vs 0.32%). Conclusion average provides estimates SIENA, it significantly more robust, probably due its built-in distortion correction. min makes suitable clinical routine. This can pave way widespread use estimation intra-scanner