Jaap van Hellemond (ErasmusMC)

Summary of presentation:
The comprehensive method to detect gastrointestinal infections is still microscopic examination of stool specimens. However, manual microscopic examination is laborious, and its accuracy is highly observer dependent. Therefore, we have developed a deep-learning approach integrated with a digital microscope scanner to detect parasites in digital images, which allows faster and more precise detection of parasites. A novel in vitro diagnostic (IVD) system (ParaScout) was developed that consists of an affordable, commercially available microscope scanner and a cloud-based deep-learning algorithm for accurate identification of helminths in stool. The performance of the ParaScout system was compared to manual examination by 4 expert technicians, using 50 validated stool-specimens containing either no parasites or one or more of the 15 helminth species for which the ParaScout system had been trained. In the 45 positive stool specimens, 63 helminth species were present, and therefore, examination by 4 technicians could have revealed 252 helminth identifications (4*63). In total the manual examinations produced 16 false negative and 11 false positive results, resulting in a sensitivity of 93.7% and a specificity of 99.6%. When ParaScout was employed to highlight suspected structures above a threshold of 0.6 for human expert confirmation, these values increased to 98.8% and 99.9%, respectively. This study demonstrated that automated scanning in combination with a deep-learning algorithm can detect helminths in stool samples with high sensitivity, while resulting in minimal false positive detections if combined with expert confirmation. This approach shows high promise for automating and improving the quality of parasite detection in patient diagnostics.