Modella AI

[airpods]

---

Members don't see this ad.

What do you think of Modella AI? It has received the breakthrough Device Designation from the FDA. Here's a video:



Even if it only screens out negative slides, this could greatly reduce the number of pathologists needed. Is this the beginning of the end?

 

[Mace1370]

Will the makers of PathChat 2a accept liability when it misses carcinoma and screens a slide as normal? The narrator said I don't have to look at normal slides.

 

[Doormat]

A couple points:

1. Path Chat 2a is a learning and teaching tool. Microscopic descriptions are useless in high-volume path labs.

2. Path Chat 2a requires digital pathology and only a small percentage of non-academic pathology labs have gone digital. Our billing company manages 32 different small and medium size pathology groups. 0 of 32 path groups have gone digital thus far. The barriers to implementation of digital pathology are huge and include: Continued requirement to make a glass slide by a histotech. High up front capital costs for the slide scanners from multiple manufacturers. Additional FTEs for personnel to load and operate slide scanners. Selection of a wide variety of whole slide imaging software companies with high annual subscription (extortion) payments to these software vendors. High cost for storage of whole slide digital images for 10 years. My observation is that only private equity, elite health systems, and academia can afford digital path. Regular health systems, pathologist-owned AP labs, and physician-owned AP labs cannot afford it.

3. Winning FDA approval for clinical application of this software is a major barrier. I estimate it will take 5 years at a minimum and each AI software package will only be approved for a single scanner type from a single manufacturer.

4. Liability issue is huge. Mace1370's post is spot on.

5. There is a ton of competition in this space. Which slide scanner manufacturer and paired software do you choose? There are so many. Roche (Ventana), PathAI, Proscia, and of course Modella all have their own software.

6. Cytology is 10X harder. A Z-stack of focal planes on a cytology smear slide makes a whole slide image 10X larger.

7. Diagnostic radiology goes down first. Digital image analysis and AI software is much further along in diagnostic radiology. Any clinical application of this AI image analysis effects diagnostic radiology a minimum of 3 years before anatomic pathology

8. Is this the beginning of the end? Possibly. It is very likely that 15-20 years from now this technology will become standard. All WSI digital images with AI software to help the pathologist do 100 cases day. However, what will the world look like in 20 years?

 

[mike-sheree]

Will the makers of PathChat 2a accept liability when it misses carcinoma and screens a slide as normal? The narrator said I don't have to look at normal slides.

The attorneys will make sure that is addressed. They are not at all concerned about sensitivity, specifity, etc. That’s our problem.

 

[HeyDalaron]

The attorneys will make sure that is addressed. They are not at all concerned about sensitivity, specifity, etc. That’s our problem.

If the corporations can prove statistically that the miss rate is the same or better than a human, then it'll go into practice.

 

[macrocyte]

Right. And if a company has a diagnostic solution that is validated to be at least as accurate as the average pathologist, then they could remove the pathologist entirely. For example, if there was a proven AI software to diagnose prostate cores or GI biopsies, the software could generate a report automatically. No need for a pathologist. And furthermore no need for all that data storage (which is one of the biggest reasons for not implementing digital pathology). The slides could be scanned and reported instantly, without keeping a large digital image. A small pic could be integrated into the report if desired. The record retained for the case would still be the glass slide. Kinda like how self driving cars aren’t keeping hours of video on file. The cameras are taking in photons and making decisions/actions in real time. Once a solution like this is approved for one specific biopsy type like prostate cores or something else, within a year many other biopsy types would follow. I used to think this type of scenario was more than a decade away, but with all the recent advances in compute and neural networks I’m not so sure.

 

[KeratinPearls]

Not a molecular pathologist but I think the reports they sign out can be done by any AI software?

 

[KeratinPearls]

Diagnexia Launches Enterprise Platform for Digital Pathology, Revolutionizing US Laboratory Coverage – Pathology News

Diagnexia is helping labs and hospitals that are in need of locums help by digitizing slides and having them read for a fraction of the cost of a typical locums. Say goodbye to locums jobs. The work from a $2000 a day locums jobs now will be available to any pathologist across the country to do at a fraction of the cost.

I have a bad feeling about this. Slides will be scanned and available to be read to the lowest bidder anywhere in the country.

I am thinking digital pathology will have a negative effect on the job market as well. Instead of hiring a full time pathologist on site, employers can just outsource the job to anyone in the country on a per slide rate.

“US laboratories are facing challenges with rising locum costs and growing coverage gaps. Our platform eliminates these constraints, delivering immediate access to qualified pathologists at a fraction of traditional costs,”

“Our technology eliminates traditional coverage constraints while enabling immediate access to US board-certified pathologists at significantly reduced costs.”

 

[ccous]

locums help by digitizing slides and having them read for a fraction of the cost

This is a concern because there will always be someone in path willing to sell themselves out and underbid for their services, but a lot of locums postings I see involve frozens and cyto onsites. So at least for those it makes more sense to have someone physically there. The need for frozens and ROSEs will keep us physically grounded in labs for a good while.

 

[caffeinegirl]

There are many pathology "AI-assistants" out there which are in practice for daily diagnostic use. Many of them are used to triage and write preliminary reports. Think of them as AI residents, and you're the attending who signs them out. It's a lot faster, and depending on the product, highly accurate. Most of the products out there are for prostate and derm, but there are also breast and lung models too.
Here's one such model from PathologyWatch which was released a few years ago.

High-fidelity detection, subtyping, and localization of five skin neoplasms using supervised and semi-supervised learning - PubMed

Our AI model, developed in collaboration with dermatopathologists, detects 5 skin lesion types with higher sensitivity than previously published AI models, and provides end users with information such as subtyping, localization, and margin status in a front-end digital display. Our end-to-end...

pubmed.ncbi.nlm.nih.gov

 

[ccous]

in practice for daily diagnostic use

I’m aware many tools have been developed but I was unaware any have been FDA cleared and are being used in anyone’s daily practice.

 

[caffeinegirl]

I’m aware many tools have been developed but I was unaware any have been FDA cleared and are being used in anyone’s daily practice.

You don't need FDA clearance if it's being used as an adjunct screening tool. But it does need to go through LDT validation.