Deep learning model screens water safety in seconds without pathogen incubation

A new artificial intelligence tool called DeepScope can classify microscopic images of unincubated water samples in seconds, offering an estimated per-test cost of $0.44 and eliminating the 24–72 hour pathogen incubation period required by conventional microbial water safety tests, which typically cost $20–$50 per test. According to the study, published in Applied Water Science, the approach reduces testing time by over 98% while meeting UNICEF’s ideal Target Product Profile numerical performance thresholds for presence/absence testing.

Fecal-contaminated water causes diseases and even death, and the delay inherent in current testing methods leaves communities waiting days for results. The DeepScope system addresses this bottleneck by using deep learning to interpret microscope images directly, bypassing the need to culture pathogens before a determination can be made. The researchers report that the model was deployed in a web server environment with mobile applications for Android and iOS, validating Internet-based and smartphone-based water safety testing with results obtained in seconds.

Building the system required assembling a dataset of microscope images of bacteria and water samples. The team drew bacterial images from the public DIBaS bacterial image collection and combined them with water samples from multiple source types. The labels were cross-validated using color-change coliform tests.

To expand dataset diversity and mitigate overfitting, the researchers developed a novel tile-swap augmentation technique. The method swaps random tile pairs within images, preserving bacterial morphology while producing a large space of augmented images. This expands the effective diversity of the training data without altering the diagnostic features the model must learn to recognize.

Four convolutional neural network models were developed using transfer learning, built on ResNet-18 and ResNet-50 backbones, each with and without dropout regularization. The best-performing configuration was the ResNet-50 model with dropout.

The model was evaluated on a field-test dataset comprising 100,000 microscope images of unseen, real-world water samples collected from fourteen different water sources across Sammamish, Washington. On the 160 original microscope acquisitions underlying this dataset, the DeepScope model achieved 96.88% accuracy, 98.61% recall, and 94.67% precision, corresponding to a 1.39% false-negative rate and a 4.55% false-positive rate. These figures describe performance on the original acquisitions; the larger 100,000-image dataset was derived from them through augmentation.

In the larger field test, conducted on the 100,000 augmented images derived from those acquisitions, the operating threshold was set to fix precision at 90%. Under these conditions the model achieved recall of 94.48% ±0.11%, accuracy of 92.79% ±0.08%, and F1 score of 92.19% ±0.09%. The false-negative rate of 5.52% falls well below UNICEF’s 10% ceiling, and the researchers describe the performance as competitive with membrane filtration and color-change coliform tests, without requiring pathogen incubation.

The study is presented as an open access article, published on 15 October 2026, and the version shared is subject to further edits before the final Version of Record. The reported results come from samples collected in a single geographic area, Sammamish, Washington, so performance across other regions, water chemistries, and contaminant profiles remains to be demonstrated. The composite water-safety determination is based on image classification rather than direct pathogen culture, and the field-test figures were obtained on augmented images derived from a limited number of original acquisitions, which the authors report alongside the original-acquisition metrics.

The system combines an estimated per-test cost of $0.44, results obtained in seconds, and deployment through a web server environment with mobile applications for Android and iOS. The researchers report that the testing validated Internet-based and smartphone-based water safety testing, and the model’s recall and precision figures met the numerical thresholds set out in UNICEF’s ideal Target Product Profile for presence/absence testing. The study reports that deep learning inference on unincubated samples achieved performance the researchers describe as competitive with membrane filtration and color-change coliform tests, without requiring pathogen incubation.

Subject of Research: Deep learning classification of microscopic images of unincubated water samples for rapid water safety testing

Article Title: A novel approach for testing water safety using deep learning inference of microscopic images of unincubated water samples

Article References: Srinivasan, S. (2026). A novel approach for testing water safety using deep learning inference of microscopic images of unincubated water samples. Applied Water Science. https://doi.org/10.1007/s13201-026-02974-z

Image Credits: AI Generated

DOI: 10.1007/s13201-026-02974-z

Keywords: water safety, deep learning, microscopy, fecal contamination, coliform testing, convolutional neural networks, ResNet, data augmentation, UNICEF, water testing, transfer learning, public health