Turning Large Electron Microscopy Datasets into Quantitative Measurements with Model-Assisted Analysis

Abstract Large electron microscopy datasets require extensive annotation and review before they can be converted into reliable measurements. ACORN is an open-source, locally executable application that offers image preparation, reviewed annotation, task-specific model training, batch inference, calibrated measurement, and statistical analysis in a single format. Users can create and correct annotations manually or with promptable segmentation and use the reviewed annotations to train models directly within ACORN. Its Cryo-EM Lab Utility (CLU) assistant also allows users to request any analysis or training operations in plain language. To show its use cases, we evaluated ACORN using low-dose cryogenic TEM images of polymeric nanoparticles and SEM images of bacterial spores. YOLO models trained in ACORN produced validation F1 scores of \(0.906\pm0.013\) and \(0.624\pm0.016\), respectively. Controlled simulations built in ACORN provided 25,983 exact object annotations for further model development and a controlled evaluation of software performance. On held-out simulation images, equivalent-circular-diameter mean absolute error was 3.60~nm (10.1\%) for TEM particles and 31.48~nm (3.7\%) for SEM spores among matched detections. The software logs every interaction, training step, and annotation in a sample history for provenance tracking. ACORN provides an environment where annotations, image calibration, model checkpoints, object measurements, and statistical records are easily accessible and traceable. This software provides a traceable path from electron microscopy images to quantitative results in multiple user-friendly formats.