Water Testing with AI

Fluid Suspended Particle Classifier

We offer US patent 11,125,675 which issued 9/21/2021 . The patent applies AI image object detection technology to locate and classify particles in a flowing fluid. For water, examples of ‘particles’ are:

  • Bacteria
  • Algae
  • Cyanobacteria
  • Diatoms
  • Dinoflagellates
  • Fish
  • Pollen
  • Larvae such as mosquito
  • Cells in blood or similar fluids
  • Inorganics

Many such species can be harmful to humans and other life.

The patent covers real-time detection, concentration estimation, and alarms as well as efficient type-specific training. The particles are located in ‘regions of interest’ which allows many advanced features. The fluid is flowing, hence mounting to microscope slides is not needed and continuous operation is possible, such as when unattended or remote.





    Dolicospermum circinale Cyanobacteria

    Neural Network AI Systems

    Artificial Intelligence systems can now identify particular ‘Regions of Interest’ or ROIs within an image, such as rectangular areas, and can classify them as particular predefined types of things. These AIs are called ‘convolutional neural network object detectors’. There can be hundreds of classes, corresponding to individual species or groups of species, depending on how the network is trained. Given thousands of images showing particular species or general types of species, the network can be trained to be more and more aware of specific characteristics of each class, so it can correctly place the ROIs into appropriate classes.

    Neural Network Output

    The output of the neural network is a set of ROIs located in the image, and for each ROI, a set of classes with numeric likelihoods for each of the candidate classes. Sometimes an ROI is clearly related to a certain class, and the likelihood for that class will be large, from to 0.5 to nearly 1.0. In that case the other candidate classes will have smaller likelihood numbers, so that the total is 1.0. Other times, an ROI may have only smaller likelihoods, with none dominating, and the result is not very clear. A well-trained network will tend to provide clearer classifications when the ROIs are actually well differentiable. There are normally some good and some poor matches in any particular set of input images.

    Neural Nets in Water Testing

    IntelliWaterAI applies neural net technology to locate particles in flowing water or other fluids, classify them, and analyze the results, producing alerts, concentration estimations, clipped images, and other information. The analysis happens in real time in a flowing fluid, with the sensor collecting the results in storage. The ROIs are used to focus the results down to the level of particles rather than to try to classify the entire image, so the analysis can be precise. Images having multiple particles and especially images with particles of mixed classes provide meaningful results. Images having multiple classes provide multiple classifications.

    Iterative Training Method

    The patent includes claims for using ‘spiked’ fluid samples to produce training images in large quantities easily. A spiked sample is one containing particles of a known class or classes. Also, successive training runs can be used to iteratively improve precision by using less specific classifiers to provide training images that are used to train more specific classifiers.

    For example, a classifier or other method that can only locate particles in general but not classify them can provide the ROIs necessary for training a classifier on a spiked sample having a single known particle class. Then, the trained classifier can be used to select classes and create ROIs from a more complex sample for further training, iteratively.

    For example, a classifier that can distinguish cyanobacteria in general can be applied to a water sample known to contain only a single cyanobacteria species, even though the sample also contains a wild mixture of particles of other classes in various concentrations, thereby creating multiple more specifically labeled training images in bulk. The sample, such as 1 Liter of collected stream water, could be used by passing it through a ‘flow cell’ on a microscope stage, and taking many images of it to be classified by the less specific classifier. Those images and the output of the less specific classifier are input to train the more specific classifier on the known classes including the known cyanobacteria. Another example could be a general cyst classifier used to generate training data to differentiate Giardia, Cryptosporidium, or other types given a water sample known to contain a single species, or at least mostly a single species.

    Applications

    By counting ROIs having similar classes in multiple images, concentrations can be estimated and alerts can be generated. The results can optionally be radioed back via channels like 4G to a central server as they become relevant, and users can remotely navigate real-time and stored clipped images to help determine species more precisely and to determine the relevance of alerts. Histories of concentrations and other measurements can be collected and displayed graphically, determining fine-grain patterns in time and space.

    Uses

    The scope of the market for the fluid suspended particle classifier is wide. There are uses such as the following:

    • Environmental health of streams, lakes, and oceans can be monitored, by checking for expected species of algae and other microorganisms
    • Surface water safety for drinking, swimming, and fishing can be ensured by detecting early-warning signs of sewage pollution such as coliform bacteria
    • Water that may contain pathogens such as giardia and cryptosporidium cysts can be monitored
    • Infections of aquatic organisms such as fish can be detected quickly when the pathogens become dispersed in farms or hatcheries
    • Larger ‘particles’ such as fish or mosquito larvae can be detected and counted.
    • Dangerous ‘Harmful Algae Blooms’ such as ‘red tide’ caused by cyanobacteria can be detected and monitored
    • Well and tank water can be monitored for contamination and growth of algae and other organisms
    • Pools can be monitored for bacteria, cryptosporidium cysts, giardia cysts,, and algae
    • Drinking water systems can do ‘in-network monitoring’ or INM to see problems with contamination in distribution lines resulting from temporary pressure loss or reversed flow.
    • Cells in blood or other fluids can be detected and counted
    • Fluids flowing across a microscope stage can be analyzed and used for training classifiers or for easy but detailed possibly dilute, possibly complex sample evaluation
    • Other fluids, such as oil, solvents, and so on can be monitored and analyzed for particles
    • Scientific, Medical, and Industrial research in general can be aided

    Partner

    Join us as an outside company or via a new company we form.

    License

    Develop a system yourself, licensing exclusively or non-exclusively from us

    Use

    Use products we develop using the technology

    Some Cyanobacteria cause Harmful Algae Blooms

    Algae blooms can affect saltwater or freshwater, and can create various toxins that may last days to weeks, even after the bloom subsides. Blooms may last for days. They cause green, reddish, or brown colored water, possibly forming scums or mats on the surface or bottom. They are more common as water temperatures rise, and as N and P concentrations increase, such as through fertilizer runoff.

    Morphological diversity in cyanobacteria.

    Rippka R, Deruelles J, Waterbury JB, Herdman M, Stanier RY (1979) Generic assignments, strain histories and properties of pure cultures of cyanobacteria. J Gen Microbiol 111:1–61.

    Cyanobacteria suspended in flowing water

    Berkshire Community College Bioscience Image Library, CC0, via Wikimedia Commons

    Coliform Bacteria are Pollution Indicators

    Coliform bacteria in water supplies or surface water are indicators of dangerous sewage pollution and are regularly tested for, such as on a weekly basis. They are rod-shaped, so they can be initially distinguished from other bacteria visually, although laboratory testing is necessary for final identification, with E. coli presence being considered conclusive. Because coliform are only indicators, they are not necessarily the only possible pathogens present. Associated pathogens include viruses, bacteria, protozoa, and parasites.

    Current testing requires many manual steps and is not real-time, including multi-day culturing on certain media to determine the ability to ferment lactose at 37°C followed by visual evaluation. Such manual testing can distinguish coliforms from naturally occurring rod-shaped bacteria, while the IntelliwaterAI system only observes morphology and size. However, IntelliWaterAI testing can be real-time and remote, in multiple locations, providing early warnings of problems, and allowing the sources to be located. Also, IntelliwaterAI testing can accumulate histories that can identify the temporary presence of or changes in concentration of rod-shaped bacteria, and that can account for a background level of naturally-occurring rod-shaped bacteria. Comparing histories at multiple locations provides important information. In the lab, IntelliWaterAI can do fast initial triage of raw samples.

    Below is a photomicrograph of some rod-shaped bacteria. Some appear end-on, and will be ignored, but others are clearly rod shaped.

    Lactobacillus bacteria, similar to E. coli, contrast enhanced

    Giardia and Other Cysts

    Giardia lamblia and Cryptosporidium form cysts about 10μm in diameter that are very infectious in humans as well as cats, dogs, and other animals. They are endemic in many areas, and are found in surface water – clear or turbid – and even swimming pools. Giardia is difficult to culture, and in fact Cryptosporidium cannot be cultured. Cryptosporidium causes ‘crypto’ and can be difficult to eradicate, being resistant to Chlorine. Neither one is routinely tested for. Here is a micrograph of a suspected Giardia cyst, and below is a diagram showing the life cycle and general appearance of Giardia.

    Suspected Giardia cyst, about 10μm suspended, unstained (Copyright © Roger L Deran)

    Ars longa, vita brevis, occasio praeceps, experimentum periculosum, judicium difficile & procrastination is expensive.

    Copyright © 2021 Roger L Deran, all rights reserved.

    Dolichospermum circinale cyanobacteria causes significant Harmful Algae Blooms