01.03.2023
Automatic brain volume and lesion segmentation is a robust method for monitoring longitudinal MRI changes in people with progressive multiple sclerosis, Gohla et al., 2023
Kurzfassung
Das Ziel der Studie war die Bewertung von Bildartefakten in der Routinebildgebung von Multipler Sklerose (MS) sowie die Genauigkeit und Robustheit der automatisierten Hirnvolumensegmentierung und Läsionsvolumetrie bei variabler Bildqualität. 35 Patienten mit progressiver MS wurden über fünf Jahre untersucht. Etwa 22% der Bilder wiesen mittlere bis starke Artefakte auf. Doch trotz Artefakten funktionierte die automatische Segmentierung des Gehirns durch AIRAscore in der Regel gut und ermöglichte eine Erkennung beschleunigter Atrophieraten bei einem Großteil der Patienten. Diese Ergebnisse unterstreichen die Nützlichkeit der automatisierten Bildgebung zur Überwachung des MS-Krankheitsverlaufs, selbst bei suboptimaler Bildqualität.
Published in
ECR 2023 EPOS C-22390
Autoren
G. Gohla, M. Kowarik, H. Tumani, A. Abdelhak, V. Bracknies, U. Ernemann, T. Lindig, B. Bender
Purpose
Multiple sclerosis (MS) is a chronic autoimmune disease of the central nervous system that causes inflammation and damage to the myelin sheath surrounding nerve fibers. One of the key features of MS is the formation of lesions in the brain and spinal cord, which can lead to physical and cognitive disability. Automated brain volume and lesion segmentation has become an important tool for monitoring MS patients, as it allows for the quantitative assessment of lesion load and corresponding brain atrophy rates. However, image quality can be a limiting factor for both visual and software-based assessment. The aim of this study was to evaluate the occurrence of image artifacts in routine imaging and the accuracy and robustness of software-based automatic brain volume segmentation and lesion volumetry under varying image quality.
Methods and materials
The study included 35 people with progressive MS (pwPMS) with either primary (PPMS) or secondary (SPMS) PMS who underwent 3D-T1‐weighted MPRAGE and 2D sagittal and transversal T2-FLAIR-weighted imaging on one 1.5T scanner over a 5-year time span. A total of 132 timepoints with 310 sequences were retrospectively segmented using AIRAscore (v2.0.1, AIRAmed GmbH, Tübingen). All segmentations and original datasets were rated by one expert neuroradiologist regarding quality on a 4-point Likert scale (excellent quality, minor artifacts, medium artifacts, strong artifacts/not diagnostic).
Results
The evaluated imaging data showed that approximately 22% of the images had medium or strong artifacts. However, despite the presence of artifacts, automatic brain volume segmentation and lesion volumetry showed diagnostic quality without or with minimal segmentation artifacts in most cases (see Figure 1 and 2). Brain atrophy over time was also analyzed and showed an increased brain volume loss (>0.5%/year) in 24/35 patients (see Figure 3) with follow-up imaging over a period of 8 to 44 months (median 30 months). This rate of brain volume loss was higher than the 95th percentile of expected brain volume change in the examined age range [1].
Conclusion
Moderate image quality is a common finding in routine imaging in pwPMS. Despite the presence of artifacts, automatic segmentation of the brain usually performs well and detects faster atrophy rates in 68.5% of pwPMS in this cohort. The correlation of reported accelerated atrophy rate with the clinical outcome in the examined population is currently being evaluated. These findings highlight the potential of automated brain volume and lesion segmentation as a valuable tool for monitoring disease progression in MS patients, even in cases of suboptimal image quality.