Journal of Scientific Research and Reports

Manual structural tracing and cohort stratification for Primary Progressive Aphasia (PPA) variants and associated Tau pathology represent a significant, error-prone bottleneck in translational neuroscience and pharmaceutical research and development (R&D). This study introduces an automated, open-source computational pipeline designed to streamline multimodal neuroimaging ingestion and candidate screening. Built upon the NiBabel and Project MONAI frameworks, the pipeline ingests raw 3D T1-weighted magnetic resonance imaging (MRI) scans in NIfTI format, processes them through dictionary-based spatial transforms, and deploys a 3D UNet architecture to automate the volumetric segmentation of target cortical subregions, specifically the left temporal lobe and the inferior parietal cortex (PC). Segmented regional volumes are normalised against total intracranial volume to engineer a localised Atrophy Index. This structural metric is subsequently fused with simulated multimodal Tau-positron emission tomography (PET) tracer uptake scores. Finally, an unsupervised learning layer using a K-Means clustering algorithm automatically stratifies subjects into distinct clinical cohorts: Mild/Stable Atrophy, Tau-Dominant Lateralized Atrophy, and Severe Multi-focal Pathology. As a foundational proof-of-concept evaluated on controlled simulated cohorts (n = 30 ), the pipeline achieved a mean Dice similarity coefficient of  0.87+- 0.03for regional boundary overlap, reducing typical multi-hour manual tracing workflows to under 45 seconds per patient scan. While these findings support computational feasibility, future work must focus on validation using real-world clinical datasets. The complete pipeline architecture, environment configurations, and source scripts are made openly accessible to support computational reproducibility and accelerate biomarker-driven candidate selection in early-phase neurodegenerative clinical trials.