Early and accurate diagnosis of Primary Ciliary Dyskinesia (PCD) allows appropriate multidisciplinary management and a reduction in lung function decline. Transmission Electron Microscopy (TEM) is essential in determining ciliary ultrastructural defects, when diagnosing PCD. This requires highly skilled specialists with considerable experience. Machine learning provides an excellent opportunity to reduce the time experts spend assessing cilia (1?2 hours) and improve accuracy of diagnosis.

In collaboration with Intel®, we have used an Artificial Intelligence platform (Intel® Geti?), to develop a workflow called PCD-AID (PCD- Artificial Intelligence Diagnosis) that uses computer vision to aid in the diagnosis of PCD. This work is part of an organised ERS Clinical Research Collaboration with BEAT-PCD. The system was tested alongside the PCD diagnostic pathway (n=158) to determine diagnostic accuracy.

The model has been trained with TEM images from over 21,000 cilia cross-sections to detect cilia and then classify them based on normal or abnormal ultrastructure or ?unusable? for diagnostic purposes (tilted or distorted images). Using retrospective and prospective patient samples, we have found PCD-AID can reliably identify ciliary ultrastructural defects (sensitivity of 0.87 and specificity of 0.88) and assess TEM images in under 1 minute per patient. It has good agreement with diagnostic specialists (> 75%) at identifying a range of ultrastructural defects and strikingly outperforms specialists at identifying subtle central pair defects associated with pathogenic mutations in HYDIN.

Implementing computer vision artificial intelligence in the diagnostic pathway improved diagnosis of PCD.