Expanding access to water quality monitoring with the open-source WaterScope testing platform

The study addresses the global challenge that approximately one in four people lack access to safely managed drinking water, with diarrhoeal diseases causing about 1 million preventable deaths annually due to faecal contamination. The World Health Organization requires that E. coli must not be detectable in any 100 ml drinking water sample, and UNICEF’s target product profile calls for testing solutions that are sensitive, rapid, low-cost, portable, robust, easy to use, and environmentally friendly. Existing methods (e.g., multiple-tube fermentation, chromogenic membrane filtration) and novel biosensing or molecular approaches often fall short for low-resource settings because of complexity, cost, logistics, lack of digitalization, and training requirements. The research question is whether an open-source, portable, digitally integrated membrane-filtration-based system with automated machine learning colony classification can provide accuracy equivalent to established methods while improving usability, robustness, cost-effectiveness, and data management in low-resource settings. The purpose is to design, validate, and field-test the WaterScope platform using human-centered design to overcome practical barriers and expand access to reliable water quality monitoring.

The paper reviews conventional E. coli monitoring methods, notably membrane filtration with chromogenic media (ISO 9308-1), most probable number approaches (ISO 9308-2/Colilert), and multiple-tube fermentation. It also surveys innovations such as biosensors, bacteriophage-based detection, PCR and other DNA amplification methods, and flow cytometry, citing reviews and comparative studies. Despite technical advances, these methods often remain unsuitable for low-resource contexts due to high costs, logistics, training burdens, and limited digital workflows. Human-centered design is highlighted as critical for overcoming WASH sector barriers, with open-source designs facilitating repair, local manufacturing, and reduced dependence on proprietary consumables.

System design and operation: WaterScope (WS) is a portable, open-source membrane filtration platform built around a reusable cartridge and single-use pre-sterilised membrane slider. The cartridge centralizes filtration, media addition, incubation, and imaging, reducing steps and contamination risk. The kit includes a combined incubator/imager with controlled illumination (LED ring), a 5 MP OV5647 camera with motorized focus, and condensation control, all housed in robust 3D-printed enclosures. Incubation uses vacuum flasks, heating pads, and a PID-controlled ATmega 328P microcontroller. Imaging and on-device ML analysis run on a Raspberry Pi (3B+/4). The case (IP67/IK08) measures approximately 29 × 23 × 26 cm and weighs ~5.8–6 kg. A diaphragm pump enables vacuum filtration through stackable funnels and a manifold for 1–2 cartridges simultaneously. The incubator holds 14 cartridges (two racks of seven). A UVC sterilizer module is included. All electronics operate at 5 V via USB-C; a 74 Wh power bank supports 21 h incubation of 14 samples. Media and cartridge: The preferred medium is CCA-like without agar (composition detailed), enabling chromogenic identification of E. coli and other coliforms. The slider holds a 13 mm absorbent pad with a 0.45 μm cellulose acetate filter. Cartridges are machined aluminium (or 3D-printed/injection-molded alternatives). Operational workflow: Five key steps: insert slider (filtration position), filter target volume, switch slider to incubation position and add ~300 μL media, incubate (8 h for rapid presence indication or 21 h for fully quantified results), and image/analyse (2–3 min) with results viewable/editable on a web dashboard or Android app. Results can be uploaded via WiFi, cellular IoT, or Android app. Sterilisation: Pre-excursion sterilization includes steam/boiling of cartridges (~1 h) and sterilization of dry media/sliders (ethylene oxide or gamma; alternative microwave-boil cycles). Field sterilization uses 80% ethanol for hands and components, careful aseptic handling, and 1 min UVC exposure for funnels/cups between samples. Automated image analysis: Images undergo cropping (Hough transform), CFU detection using a YOLOv8-based model trained on >30,000 expert-labelled CFUs, classification of colonies by HSV colour profile (E. coli, other coliforms, premature), outlier detection via k-means on dominant colours to flag TNTC/sediment-laden images, and manual review/edit on the dashboard (VIA tool). The ML runs locally on the Raspberry Pi without internet. Human-centered design process: Iterative workshops (India, Tanzania, South Sudan; 75 participants) collected usability feedback leading to design changes such as adding an electric vacuum pump to reduce effort and doubling incubator capacity to 14 samples. Validation study designs: Three validation tiers were used. 1) Controlled laboratory experiments following ISO 17994 principles used a five-step dilution series targeting ~10, 20, 40, 80, 160 CFU/100 ml; methods compared: WS, CCA (ISO 9308-1), Colilert-18 MPN (ISO 9308-2), and MLSB (as in DelAgua kits). Fourteen replicates per dilution per method; negative controls (autoclaved E. coli) included. 2) Controlled environmental validation on the River Cam (UK) biweekly in 2022 compared WS, CCA, and Colilert with 2–5 replicates each sampling date; volumes adjusted based on turbidity to ensure countable plates. 3) International field trials in Juba (South Sudan, May 2022), Kawangware (Kenya, Sep 2022), and Addis Ababa (Ethiopia, Jun 2023): 301 WS samples total, each paired with Compact Dry as reference; iterative procedural improvements (sterilization training emphasizing 80% ethanol; addition of UVC funnels sterilization) between trials. For studies 2–3, volume corrections were applied for CFU/100 ml based on processed volumes. Counts from non-WS methods followed manufacturer protocols. Statistical analyses: Linear regression for method linearity, mean-difference equivalence testing broadly following ISO 17994 using 95% CI on differences in log10 mean counts, and two-sided Spearman rank correlations. For field trials, ROC curves and AUCs computed by varying detection thresholds (1–20 CFU/100 ml) using Compact Dry as reference. Duplicate/repeat pairings were randomized to avoid bias.

• Lab validation (controlled): Strong linear relationships between WS and references with R2 of 0.92 (WS vs CCA), 0.84 (WS vs Colilert), and 0.75 (WS vs MLSB). No significant differences at 95% confidence between WS and any reference method by mean-difference analysis. Spearman correlations: WS with CCA/Colilert 0.94/0.93, similar to CCA vs Colilert (0.95). Negative controls showed no growth.
• River Cam validation (environmental): Seasonal EC variation observed across all methods. WS showed R2 of 0.95 (vs CCA) and 0.94 (vs Colilert); Spearman 0.98 and 0.94 respectively. Mean-difference analyses indicated no significant differences at 95% confidence. Negative controls showed no growth.
• International field trials: 301 WS samples across Juba, Kawangware, Addis Ababa, referenced to Compact Dry. Overall strong correlation with Compact Dry; initial false positives in Juba traced to residual contamination in reusable funnels. Training emphasizing 80% ethanol (vs 100%) plus introduction of UVC sterilization reduced FP rate to 5% in Addis Ababa. Spearman rank across all trials: 0.75 including FPs, 0.91 excluding FPs. ROC AUC improved across successive trials (exact AUCs shown in paper’s Fig. 4c).
• Early detection capability: In Juba and Kawangware field data (n=130), 67% of environmental samples with contamination could be reliably detected within 8 h, enabling faster response.
• Usability: PSSUQ-based assessment showed favourable usability for WS, especially after hands-on training; participants preferred WS for accuracy, shorter hands-on time, easier teaching, and trustworthiness.
• Automated counting: Strong correlation between automated ML counts and manual counts, supporting reliable automated classification.
• Cost analysis: WS kit priced at $1499 and $2.95 per test. Over a two-year campaign, WS offers approximately 26% cost savings relative to an equivalent WagTech portable kit with Nutridisks for E. coli specificity. WS has similar overall cost to CompactDry (about $0.25 more per test when logistics, labour, and equipment are included). Open-source local production could further reduce costs.
• System capacity and robustness: Portable case ~29 × 23 × 26 cm, ~5.8–6 kg; 14-sample capacity; runs on 5 V USB-C power with a 74 Wh power bank sufficient for 14 samples with 21 h incubation. Integrated data capture and cloud/IoT/App connectivity enable rapid reporting and traceability.

The findings demonstrate that the WaterScope platform achieves microbiological equivalence to standard reference methods (ISO 9308-1 CCA and ISO 9308-2 Colilert) in both controlled laboratory and environmental river settings, addressing the core research goal of delivering accurate E. coli quantification in a portable, low-resource-appropriate system. The integration of controlled imaging, tailored ML classification trained on a large labelled dataset, and simplified cartridge/slider workflow reduces procedural complexity and training demands while enabling digital data capture and rapid dissemination. Field trials revealed real-world challenges (e.g., sterilization protocol adherence) that were mitigated through updated training and UVC sterilization, improving specificity and reducing false positives to low levels. Usability assessments support the human-centered design approach, indicating that structured training further enhances user experience and reliability. Cost analyses suggest the platform can reduce monitoring expenses relative to established kits, particularly when leveraging open-source local manufacturing pathways. Collectively, the results indicate that WaterScope can strengthen WASH monitoring programs by enabling accurate, timely, and traceable water quality data collection in resource-constrained environments.

WaterScope is an open-source, portable, and digitally integrated membrane filtration platform that provides accurate E. coli quantification equivalent to established methods while improving usability, robustness, and data management. Validations across laboratory, river, and multi-country field contexts confirm strong agreement with reference standards and demonstrate effective mitigation of field-specific challenges through iterative design and training. The system’s modular cartridge/slider design and embedded imaging/ML analysis streamline workflows and enable broader analytical capabilities. Future work includes expanding validated applications to colorimetric assays (e.g., chlorine, metals, arsenic), clinical testing (e.g., urine analysis and antibiotic susceptibility), and integrating the dashboard with public health and water monitoring systems such as SORMAS and mWater. As an open-source platform, community-driven improvements in build processes, training, and cost reduction are anticipated to further enhance scalability and impact toward global water equality.

• Initial field deployment exhibited false positives due to residual contamination of reusable funnels; resolved through improved training (emphasis on 80% ethanol) and addition of UVC sterilization, but underscores the importance of rigorous sterilization in routine use.
• Users without prior WASH experience found self-guided training materials less detailed than conventional MF methods; hands-on training alleviated this, indicating training materials remain a critical factor for adoption.
• At higher colony densities (approaching or exceeding ~160 CFU/100 ml), colony overlap can make accurate counting more challenging; sample volume adjustment and turbidity-based volume selection were used to mitigate this.
• Field validations, while spanning three locations in East Africa and multiple water source types, are geographically limited; broader multi-region evaluations and long-term deployments would further support generalizability.