Mobile microscopy and telemedicine platform assisted by deep learning for quantification ofTrichuris trichiurainfection

Abstract

Soil-transmitted helminths (STH) are the most prevalent pathogens among the group of neglected tropical diseases (NTDs). Kato-Katz technique is the diagnosis method recommended by WHO and although is generally more sensitive than other microscopic methods in high transmission settings, it often presents a decreased sensitivity in low transmission settings and it is labour intensive. Digitizing the samples could provide a solution which allows to store the samples in a digital database and perform remote analysis. Artificial intelligence methods based on digitized samples can support diagnostics efforts by support diagnostics efforts by performing an automatic and objective quantification of disease infection.In this work, we propose an end-to-end pipeline for microscopy image digitization and automatic analysis of digitized images of soil-transmitted helminths. Our solution includes (1) a digitalization system based on a mobile app that digitizes the microscope samples using a low-cost 3D-printed microscope adapter, (2) a telemedicine platform for remote analysis and labelling and (3) novel deep learning algorithms for automatic assessment and quantification of parasitological infection of STH.This work has been evaluated by comparing the STH quantification using both a manual remote analysis based on the digitized images and the AI-assisted quantification against the reference method based on conventional microscopy. The deep learning algorithm has been trained and tested on 41 slides of stool samples containing 949 eggs from 6 different subjects using a cross-validation strategy obtaining a mean precision of 98,44% and mean recall of 80,94%. The results also proved the potential of generalization capability of the method at identifying different types of helminth eggs.In conclusion, this work has presented a comprehensive pipeline using smartphone-based microscopy integrated with a telemedicine platform for automatic image analysis and quantification of STH infection using artificial intelligence models.

Affordable artificial intelligence-based digital pathology for neglected tropical diseases: A proof-of-concept for the detection of soil-transmitted helminths and Schistosoma mansoni eggs in Kato-Katz stool thick smears

“…The lower precision (91.8%) and lower recall (86.2%) for SCH can largely be attributed to the limited size of the training data collected. The performance of our AI model for STH and SCH is comparable to or better than previously published AI based approaches [ 15 , 17 , 18 , 28 , 31 – 35 ].…”

“…Although the outages may last only seconds or minutes, the power interruptions also interrupted the scanning process and resulted in the loss of data for the current scan. While some other WSI approaches require cloud-based operations [ 15 – 17 ], access to the internet was very limited and unreliable in the remote endemic areas where our prototype systems were evaluated. Potential solutions for this first challenge would be to reduce the size of the WSI scanner to enable more scanners per equipment case and to integrate or include an uninterruptible power supply (UPS) with battery backup with the system.…”

“…The collection of over 1.3 million images containing 16,990 verified egg annotations across four helminth species was collected from four geographic regions with our prototype WSI scanner, which provided added variability in the stool appearance and provided valuable and practical lessons in how AI-DP solutions can fit into field settings. As a comparator to another KK image collection, the Kankanet approach used a dataset of 2,078 images and the unseen dataset was based on 185 egg images [ 18 ], or a test set (15%) that consisted of 136 Trichuris and 647 Ascaris egg images [ 17 ]. Other AI models that were not trained and evaluated with images collected from KK stool thick smears are likely not transferable to KK images and are likely to experience a significant reduction in precision and recall due to the additional egg-like artefacts and debris present in KK stool thick smears.…”