Validation of Digital Slide Scanning and a Convolutional Neural Network for the Detection of Intestinal Parasites in...

Area of Science:

• Clinical Microbiology
• Parasitology
• Digital Pathology
• Artificial Intelligence in Diagnostics

Background:

• Digital microscopy (DM) combined with convolutional neural network (CNN) models shows promise for detecting intestinal parasites in stool samples.
• Previous studies validated DM/CNN for detecting protozoa and helminth ova/larvae in various staining methods.
• This study aimed to evaluate the diagnostic performance of a specific DM/CNN workflow in a routine clinical microbiology setting.

Purpose of the Study:

• To assess the diagnostic accuracy and reliability of a digital microscopy and convolutional neural network workflow for routine intestinal parasite detection.
• To compare the performance of the DM/CNN system against traditional light microscopy in a clinical laboratory environment.
• To evaluate the potential of this technology to assist laboratory technicians and streamline the parasitological workflow.

Main Methods:

• A clinical validation study utilized the Grundium Ocus 40 scanner and the Techcyte Human Fecal Wet Mount (HFW) algorithm.
• The system was tested on 135 reference samples and 208 routine clinical samples.
• Analytical sensitivity, precision, limit of detection (LOD), and agreement with light microscopy (LM) were assessed.

Main Results:

• The DM/CNN workflow demonstrated high slide-level agreement with LM: 97.6% positive and 96.0% negative for reference samples.
• Prospective testing on routine samples showed an overall agreement of 98.1% (Cohen's Kappa = 0.915) between DM/CNN and LM.
• The system exhibited high reproducibility and stability, though analytical sensitivity slightly decreased at higher dilutions; minor discrepancies noted for *Blastocystis* spp. with higher DM/CNN sensitivity.

Conclusions:

• The Grundium Ocus 40 scanner and Techcyte HFW algorithm combination provides a reliable, low-throughput screening solution for intestinal parasite detection.
• The DM/CNN workflow effectively assists diagnostic technicians by pre-classifying potential parasitic structures, reducing manual review time and simplifying the workflow.
• Successful implementation requires site-specific validation, optimization of confidence thresholds, and consideration of sample processing and imaging variations.