PixelVision: A Modular Web-Based Framework for AI-Assisted Biological Image Analysis Across Diverse Microscopy Modalities

The increasing volume and heterogeneity of microscopy data in biomedical research present challenges for automated image analysis and the integration of assay-specific computational methods. We developed PixelVision, a modular, web-based framework that integrates deep learning models, quantitative image analysis, and expert review across diverse biological imaging modalities. The framework supports single-organism imaging, two- and three-dimensional cell culture analysis, whole-slide imaging, and interactive annotation through a shared infrastructure for model deployment, version management, and data processing. Quantitative evaluations demonstrated application-specific classification and segmentation performance, with inference throughput of approximately seven high-resolution images per second on a single GPU. In a prospective zebrafish screening study, iterative model refinement using expert-reviewed classifications was associated with an increase in macro-averaged F1 score from 0.867 to 0.967 across three screening rounds. PixelVision provides a framework for integrating automated image analysis and expert assessment across heterogeneous microscopy workflows.