Precise measurement of nanoscopic septin ring structures with deep learning-assisted quantitative superresolution microscopy

Abstract

Deep learning (DL)-based recognition and analysis of structures in superresolution microscopy data is prone to bias. Validation of DL models on cellular and simulated data allows for unbiased recognition of septin structures different in size from wildtype providing a new experimental system for the investigation of septin polymerization.

AI analysis of super-resolution microscopy: Biological discovery in the absence of ground truth

“…Deep learning is achieving state-of-the-art results on a wide array of prediction applications (e.g., classification and segmentation), surpassing their shallow machine learning counterparts often by big margins, and even meeting or surpassing human experts on biomedical image interpretation tasks ( Fujisawa et al, 2019 ; Rajpurkar et al, 2017 , Preprint ; Zhang et al, 2019 ). While deep learning has found rapid adoption in SRM acquisition and image generation–related tasks ( Hyun and Kim, 2023 ), discovery-oriented SRM data analysis is still limited ( Khater et al, 2019a ; Zehtabian et al, 2022 ). Explainable AI (XAI) is a fast-growing field aimed at improving our understanding of deep features and explaining deep model decision processes.…”