Research
Publications from our research division, technical writing from our engineering team, and scientific commentary on the problems we find most interesting.
Peer-Reviewed Publications
Yashara Biosciences Research Division
We present a hybrid architecture combining vision transformer backbones with convolutional upsampling decoders for cell segmentation in multi-channel fluorescence microscopy. Evaluated across 14 diverse cell types and 6 imaging modalities, the model achieves state-of-the-art segmentation accuracy while maintaining practical inference throughput on standard laboratory hardware.
Yashara Biosciences Research Division
High-content screening generates large volumes of imaging data with variable per-well quality. We describe a statistical framework for automated QC flagging that identifies illumination artifacts, focus drift, edge effects, and cell confluency deviations without manual review, reducing analyst time by 78% in a prospective evaluation across five pharmaceutical screening campaigns.
Yashara Biosciences Regulatory Research Division
We surveyed 412 clinical laboratories across 38 states to quantify the time elapsed between CLIA regulatory changes published in the Federal Register and laboratory awareness of those changes. The median latency was 47 days; 23% of surveyed laboratories were unaware of changes more than six months after publication. We describe the implications for compliance risk and propose a real-time regulatory monitoring framework.
Yashara Biosciences Research Division
Volumetric segmentation of cells in light-sheet fluorescence microscopy remains computationally demanding and methodologically inconsistent across research groups. We describe a multi-scale transformer architecture trained on a curated dataset of 1.2 million annotated 3D cell instances across eight model organisms, achieving reproducible segmentation across diverse tissue types with a processing rate compatible with routine research use.
Technical Writing
EngineeringThe case for hybrid architectures in biological image analysis — balancing global context modelling with the inductive biases that make CNNs effective on natural images.
ScienceA critical look at how the field measures segmentation accuracy, and why the metrics that dominate benchmark leaderboards can mislead real-world evaluation decisions.
ComplianceA summary of our survey of 412 clinical laboratories — and what the median 47-day regulatory awareness latency means for compliance risk.
EngineeringWhat we learned building a deep learning segmentation plugin that integrates invisibly into the ImageJ ecosystem — and the traps we fell into along the way.
ScienceThe structural differences between light-sheet and confocal fluorescence data, and why a model trained on one does not generalise well to the other without explicit architecture choices.
ComplianceWhy deficiency tags are dramatically more expensive to remediate than to prevent — and what that arithmetic means for how laboratories should allocate compliance resources.
We collaborate with academic research groups, imaging core facilities, and clinical laboratories on studies relevant to our platforms.
