CiliaScope - Transforming Respiratory Diagnostics With AI

 
Respiratory cilia are tiny, hair-like structures that line the airways and help keep the lungs healthy by clearing out mucus, dust, and other harmful particles. When these cilia do not function properly, it can lead to a rare genetic condition called Primary Ciliary Dyskinesia (PCD), which affects breathing and lung health.
 
Diagnosing PCD currently involves a slow and labour-intensive process. Specialist clinicians must manually search for cilia in extremely thin tissue samples using a transmission electron microscope (TEM). Because cilia are so small and can appear in many different orientations, clinicians must take many images at different magnifications, identify the right cross-sections, and analyse ~30-50 examples per sample to make an accurate diagnosis. Although machine learning (ML) tools are beginning to assist with diagnosis, image collection and classification are still done manually, creating a bottleneck that limits the speed and scalability of diagnosis.
 
The goal of CiliaScope is to automate the image collection and analysis process by controlling the microscope to scan low-resolution images, locate regions containing cilia, capture high-resolution images of each cilium, and generate fully reproducible diagnostic reports. The full algorithm has been developed, but at the end of our Impact Acceleration Account  project, we encountered an unexpected, high-risk and fundamental bottleneck that prevented CiliaScope from functioning as intended. TEMs do not maintain a fixed reference frame when changing magnification, which means that coordinates identified at low magnification do not precisely align with those at high magnification. Unless this problem is solved, CiliaScope cannot operate reliably. Addressing this issue is the central aim of this MRC proposal.
 
If successful, CiliaScope will dramatically reduce the time and cost of diagnosis, release skilled staff for other tasks, and enable microscopes to operate...