Data-Driven differentiation of idiopathic Normal-Pressure hydrocephalus and progressive supranuclear palsy via automated volumetric analysis.
Authors
Affiliations (10)
Affiliations (10)
- Department of Radiology, Inje University Busan Paik Hospital, Busan, Korea, Republic of.
- postech, Pohang, Korea, Republic of.
- Vuno (South Korea), Seoul, Korea, Republic of.
- Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Korea, Republic of. [email protected].
- UC Berkeley-UCSF Joint Graduate Program in Bioengineering, Berkeley, CA, United States.
- Department of Radiology, Seoul National University College of Medicine, Seoul National University Hospital, Seoul, Korea, Republic of.
- Healthcare AI Research Institute, Seoul National University Hospital, Seoul, Korea, Republic of.
- Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Korea, Republic of.
- Department of Neurology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Korea, Republic of.
- Department of Radiology, Sungkyunkwan University School of Medicine, Samsung Medical Center, Seoul, Korea, Republic of.
Abstract
This study aims to develop automated machine learning methods to differentiate idiopathic normal pressure hydrocephalus (iNPH) from progressive supranuclear palsy (PSP) using advanced imaging features and volumetric analysis. We analyzed T1-weighted 3D brain MRI scans of iNPH and PSP patients using automated component measurement acquisition and deep learning-based automated volumetric analysis. We grouped the MRI features as brainstem subgroup, volumetrics subgroup, midbrain to pons (MP) ratio subgroup (included the midbrain to pons area ratio and volume ratio), and disproportionately enlarged subarachnoid space hydrocephalus (DESH) subgroup (included the callosal angle, Sylvian fissure empty ratio, vertex region crowding ratio, and Evans' index). Key imaging features were quantified, and statistical comparisons were conducted to identify distinguishing characteristics. Machine learning models were applied to evaluate feature effectiveness and improve diagnostic classification accuracy. This study analyzed 192 patients (132 iNPH, 60 PSP) and found significant differences in midbrain volume, midbrain to pons volume ratio, callosal angle, Sylvian fissure empty ratio, vertex region crowding ratio, and Evans' index. Machine learning models, particularly the linear Support Vector Machine (SVM), achieved high diagnostic accuracy (AUROC = 0.98) in distinguishing iNPH from PSP. Volumetric analysis outperformed other feature subgroups. The deep learning-based automated brain volumetric analysis achieved high diagnostic accuracy in distinguishing iNPH from PSP using T1-weighted brain MR images.