Multiplex imaging combined to machine learning enable automated profiling of cortical malformations: applications in tuberous sclerosis complex

Malformations of cortical development such as tuberous sclerosis complex arise within a heterogeneous cellular landscape that conventional histopathology only partially resolves. Here, we combined a 19-marker multiplex immunofluorescence panel with a machine learning-driven image analysis pipeline to map and quantify over 365 000 cells from paediatric surgical cortex, defining the single-cell architecture of TSC lesions. Microtubers were objectively delineated by vimentin and detected in all TSC samples but absent from non-dysplastic controls. Within these structures, balloon cells exhibiting strong pS6 activation occupied lesion cores and were confined to microtuber boundaries, whereas dysmorphic neurons were more diffusely distributed into adjacent cortex. The microtuber niche was dominated by astroglial remodeling: immature and reactive vimentin-positive astrocytes, including Lamp5-positive subsets, accumulated at and around lesion rims, while mature GFAP-positive astrocytes showed only modest changes. Distance-based spatial analyses revealed neuronal exclusion from microtuber centres with gradual recovery in surrounding tissue, indicating local network disruption. Unsupervised clustering and niche modelling recapitulated these spatial gradients, identifying a glial-dominated ecosystem that concentrates balloon cells, increases inter-neuronal distances, and reduces cell-cell interactions. Together, these data support a model in which cortical tubers arise through the coalescence of microtubers orchestrated by balloon cells and reactive gliosis during corticogenesis. Beyond elucidating disease architecture, our automated framework enables reproducible lesion detection, quantitative cell-state mapping, and spatial readouts applicable across malformations of cortical development.