Automated Morphological Profiling via Deep Learning-Based Segmentation for High-Throughput Phenotypic Screening

This study presents a fully automated workflow for morphological profiling of microscopy images using deep learning–based segmentation built on the IKOSA platform. The system was developed to reduce the manual setup and technical expertise typically required in established tools such as CellProfiler, while maintaining compatibility with existing analysis standards. Using data from the JUMP Cell Painting pilot dataset, the workflow generates segmentation masks directly from raw microscopy images and extracts detailed cellular features for downstream analysis.

The final model demonstrated strong segmentation performance, achieving precision, recall, and average precision values of up to 0.98, 0.94, and 0.92 for nuclei segmentation. It processed full-resolution images (1080 × 1080 pixels) in approximately 2.2 seconds per image, supporting efficient large-scale analysis. From the segmentation output, 3664 morphological features were generated and reduced to 1145 key descriptors through correlation-based prioritization, enabling reliable phenotype profiling while limiting redundancy.

Overall, the IKOSA-based workflow showed high agreement with established CellProfiler measurements, with a normalized mean absolute error of 0.0298, confirming measurement consistency. By integrating automated segmentation and feature extraction into a unified pipeline, the approach reduces configuration effort and improves reproducibility, supporting scalable high-content imaging workflows in drug discovery and precision medicine.