Comparative Extraction of Cellular Features from High-Resolution Volume Imaging

Abstract

High-resolution volume imaging techniques, such as lattice light-sheet microscopy (LLSM), generate vast and complex datasets that demand advanced analytical approaches to uncover biologically meaningful insights. While LLSMs high spatial and temporal resolution provides critical data for understanding cellular processes, distinguishing subtle differences between cells in distinct states remains challenging. Here, using an adaptive, human-interpretable kernel-based calculation strategy, we developed Machine-Learning-Based Visual Extraction of Structural Features (M-VEST), a method designed to identify and interpret structural differences with high precision. By applying M-VEST to mitotic cells, we uncovered novel functions of the oncogene Aurora kinase A, demonstrating its utility in revealing previously undetected features. Validated using LLSM datasets, M-VEST offers a scalable framework for analyzing large and complex imaging data, advancing insights into cellular dynamics and beyond.