Filamentous Bacteria Identification Now Can Be Just an Email Away

A new technology helps operators get a biological analysis of the treatment process using an online tool, a microscope, and a smartphone.

Filamentous Bacteria Identification Now Can Be Just an Email Away

Opseyes enables operators to get quick identification of filaments using a microscope and a smartphone camera.

Interested in Treatment?

Get Treatment articles, news and videos right in your inbox! Sign up now.

Treatment + Get Alerts

Microbiological analysis can be a powerful tool for managing biological treatment and diagnosing and resolving process issues.

Now there’s an online technology that lets operators get a nearly instant microscopy assessment of filamentous bacteria by way of a smartphone. Opseyes has launched an artificial-intelligence-powered tool that can provide a rapid check on plant conditions and deliver expert recommendations to address potential concerns.

Developed by Ramboll, a Denmark-based consultancy that includes a specialty in water and wastewater treatment, the tool sends an analysis to the user’s inbox after online submission of a microscopy sample. The company says the ability to respond to changes in bacteria in real time helps operators consistently meet permit requirements. 

The analysis requires only a simple three-step process, according to Bryan Arndt, CEO of Opseyes. Arndt talked about the technology in an interview with Treatment Plant Operator.

Why did you develop this tool?

Arndt: I‘ve been running wastewater plants for about 20 years as an engineer and troubleshooter. Fifteen years ago, I was working at a landfill, and we had a huge filament problem. We sent a sample to a lab to get the filament identified. The plant was overflowing every day, and I was in my waders sweeping up leachate and bacteria every day for five days waiting on the lab test result. I really wanted that fixed.

How did that experience lead to the development of the technology you now offer?

Arndt: My brother told me how doctors are using artificial intelligence to read X-rays. As good as a doctor is, if you combine that expertise with AI, it’s even better. And I said, maybe if AI can read X-rays, I can teach it to read bacteria. We took the same basic technology and fed it a ton of images, and it learned to identify the bacteria.

What steps does an operator use to get the online analysis?

Arndt: Most plants have a microscope because somebody bought one at some point, but it sits in a corner and nobody knows how to use it. We provide an attachment so they can take a cellphone, attach it to the microscope, and take four different pictures from a wastewater sample. They go to our website and upload the pictures, and the AI does the work. It identifies the filaments in the sample and sends a report back.

How many filaments can your tool identify?

Arndt: Right now, we can identify six of the most common filaments.

Why can’t operators just look through a microscope to make the identification?

Arndt: Filament identification is hard to learn. You’re looking at the image and asking, is that squiggle Nocardia? Or is that squiggle S. natans? Or is that squiggle Thiothrix? I’ve taken a class personally, and I had trouble doing it. That’s part of the reason I wanted to develop this process. It’s hard to identify in the field what filaments you have, but AI can do it accurately.

What is involved in training the AI tool?

Arndt: We take, say, 10,000 images of the target filament. We set 20% of the images aside and teach the AI by feeding the remaining 80% to it. Then we feed it the rest of the images and make it identify them. 

What level of accuracy are you seeing in the filament identification?

Arndt: We are seeing at least 90% accuracy. We are having the reports reviewed by professional microbiologists as well, until we are 100% sure this tool is perfect.

How long does it take from the time the operator uploads the picture until the report is sent back?

Arndt: We send out a draft from the AI in 10 minutes or less. Then our biologist reviews it, and a final report is sent within 24 hours — generally much sooner than that.

Do the slides used to view the sample need staining or other preparation?

Arndt: No. That’s part of why AI is very useful. It can detect things that are a lot harder for humans to see. It’s pattern recognition, so we don’t need any staining at all.

Are there any special techniques for taking the sample?

Arndt: No. They just follow the standard protocol. We need a representative sample. So don’t take a sample from a corner that never gets mixed, and don’t take one that has no bacteria in it because you took it off the top of the clarifier. Just take a standard bulk sample you should be taking anyway every day.

How easy is it to connect the cellphone to the microscope?

Arndt: You simply put the phone in a U-bracket that has soft padding. Then you aim it at the microscope. It looks right down where your eye would go. Or if you have a camera connector on your microscope, you can attach the phone to that, too.

What magnification is used to view the sample and take the picture?

Arndt: We prefer 100X to 400X.

What is included in the report that operators receive?

Arndt: It tells whether we detected a filament and whether it’s a significant problem. It gives a description of the filament, what causes it, and a little background on it. If it’s a significant problem, it gives you steps to correct it. For example, it might say this filament is vulnerable to chlorination, so chlorinate your return activated sludge. If the filament is not abundant enough to cause a problem but we’re seeing a predominant type, the report will recommend things to watch so it doesn’t become a problem in the future.

How do you envision expanding this tool’s capability in microbiology analysis?

Arndt: We want to keep adding filaments. We also want to add higher life forms like rotifers, tardigrades, protozoa and stalked ciliates. That way we would be able to tell operators their sludge age and make recommendations accordingly.

What has been the response from operators who see this technology in action?

Arndt: A lot of them really like it. They know this is a good thing to do for checking out their plant. It can identify high loading, underloading, oil and grease problems, and septic conditions in the sewer network. Our client base is very receptive to the concept. I am excited to finally get to do this and help fix a problem I had years ago.   


Comments on this site are submitted by users and are not endorsed by nor do they reflect the views or opinions of COLE Publishing, Inc. Comments are moderated before being posted.