Toward Generalizable Whole Brain Representations with High-Resolution Light-Sheet Data

Unprecedented visual details of biological structures are being revealed bysubcellular-resolution whole-brain 3D microscopy data, enabled by recent advancesin intact tissue processing and light-sheet fluorescence microscopy (LSFM).These volumetric data offer rich morphological and spatial cellular information,however, the lack of scalable data processing and analysis methods tailored tothese petabyte-scale data poses a substantial challenge for accurate interpretation.Further, existing models for visual tasks such as object detection and classificationstruggle to generalize to this type of data. To accelerate the development of suitablemethods and foundational models, we present CANVAS, a comprehensive setof high-resolution whole mouse brain LSFM benchmark data, encompassing sixneuronal and immune cell-type markers, along with a set of cell annotations and aleaderboard. We also demonstrate challenges in generalization of baseline modelsbuilt on existing architectures, especially due to the heterogeneity in cellular morphologyacross phenotypes and anatomical locations in the brain. To the best ofour knowledge, CANVAS is the first and largest LSFM benchmark capturing intactmouse brain tissue at subcellular level, and includes extensive annotations of cellsthroughout the brain.