Abstract:
Digital pathology has the potential to revolutionize cancer diagnosis and treatment, particularly in resource-poor settings. This conversation highlights the development of a novel, low-cost digital pathology workflow that integrates AI-powered image analysis and remote pathology expertise. The system aims to optimize diagnostic accuracy, reduce turnaround times, and improve patient outcomes. Key challenges, including staining requirements, microvasculature analysis, and community-based screening, are discussed. The conversation emphasizes the need for innovative solutions to address the disparities in cancer diagnosis and treatment in resource-constrained environments.
*Key words:* Digital pathology, AI-powered image analysis, remote pathology expertise, resource-poor settings, cancer diagnosis, diagnostic accuracy, patient outcomes.
Conversational Transcripts:
[06/12, 08:25] NS: Happy to share news of our NIH U-01 grant for cancer technology
[06/12, 08:28] A: Congratulations Doc..👌👍
[06/12, 08:39]rb: Congratulations!👏
[06/12, 09:17]rb: Was trying to google more specifics about how the data capture is done with the device your team has prototyped. Is there a publicly available link with more details about it's workings?
[06/12, 09:24] NS: Still building the newer prototype 🙂 happy to explain
[06/12, 09:33]rb: Thanks. Wanted to understand the line below 👇
"Our prototype will digitise the images locally, so only the images need to be transmitted over the internet for analysis."
So does it mean the doctor takes the biopsy but still needs a pathology team to process the sample by creating tissue blocks and then stain and then take a microscopic digital image or does the new prototype drastically works around the usual pathology processing workflows?
[06/12, 09:42] NS: We’ve created a mechanised workflow to standardise the cytology/histopathology system to make it faster more efficient and less prone to error. AI/ML extraction of abnormal cells that are then sent over mobile networks to a pathologist remotely for interpretation
[06/12, 09:45]rb: This sounds very interesting and exciting. Would be keen to understand better how the regular cytology, histopathology workflow is optimised for better efficiency. I am guessing AI comes in image pattern recognition and filtering out noise and capturing the actual area of interest?
[06/12, 09:47] NS: That’s one aspect. The other is a low cost workflow to standardize the creation of a cytology smear and histopathology slide. That’s what we have got resources to build out, currently we are a prototype stage
[06/12, 09:49]rb: Yes what you mention here is the most interesting aspect from my own rural medical college low resource lens
[06/12, 09:51]rb: As a rural medical college physician regularly left with biopsy specimens that generate more diagnostic uncertainty than useful closure, I feel it would be great if your prototype also helps to make all these real patient data accessible online for global collective medical cognition and real time feedback
[06/12, 09:53] NS: Absolutely what we had in mind. Closing the gap in a real sense. Plenty of people like doctors to take a biopsy, no one to process the sample, create a slide to interpret. Which is why we went after the consolidated workflow and not just a fancy hardware design which in isolation will solve nothing. My experience from a lot of cancer screening camps that did not add the value intended
[06/12, 09:55] NS: By extracting abnormal cells and sending over mobile internet, doesn’t require massive file sizes usually associated with digital pathology, which is untenable without broadband connectivity
[06/12, 10:02]rb: Yes currently we do this manually when our pathologists point to an uncertain area in the slide and that particular area is filmed and shared with other human pathology practitioner networks.
What still is a challenge in our low resource settings is the inability to stain to certain requirements, for example we recently did a pathological autopsy of a woman with CD (disseminated Tuberculosis) but she also had an NCD (metabolic syn, diabetes) for years and while performing the autopsy (I'm a physician by the way not a pathologist) we found gross evidence of aortic thickening and expected some interesting findings in the coronary microvasculature but with our earlier tryst with our dental pathologists we already knew that studying the microvasculature appears to be out of bounds for most pathologists as it requires very specialised stains.
This is a current blow as most of our NCD heart failures are Hfpef (as in that particular patient) which are generally attributed to coronary microvasculopathy.
[06/12, 10:04] NS: Yup absolutely. ROI algorithms can do this very well. Digital pathology is expanding daily, the only constraint is the cost of hardware so that’s what we went after to have the biggest impact and likelihood of success
[06/12, 18:29] AJ: Congratulations Dr. Isn't screening at pre-cancer stage more effective in managing the cancer disease burden? Is biopsy the only way to screen cancers?
[06/12, 18:40]K : To do this, the sample need to be processed and slide need to be prepared and stained. Then in microscopy a pathologist need to capture abnormal area and upload in to your application then it will give possible diagnosis and future tests to be done?
[06/12, 19:15] NS: Pathologist remote and only for interpretation, rest the same
[06/12, 19:16] NS: Definitely. But not feasible without community based door to door screening which no one pays for. Opportunistic screening the only thing that works currently
[07/12, 11:32) JR: There are more options.
Will DM you soon
[07/12, 11:33] JR: Not exactly.
Better models to screen are there
06/12, 20:46] A : Indian Govt is planning on Door to Door Surveys from whatever I know and screening cancer is also a priority...
[07/12, 11:43] NS: Have been hearing this. Hope they invest in it. India pays more per oral cancer patient than other BRICS nations because early detection is rare. 75% are locally advanced cancers at diagnosis
[07/12, 11:45] D: Screening as a definition should focus on early detection and asymptomatic patients
Now, let's move on to the thematic analysis:
*Thematic Analysis:*
The conversation revolves around the theme of digital pathology and its potential to improve cancer diagnosis and treatment in resource-poor settings.
*Codes:*
1. Digital pathology
2. AI-powered image analysis
3. Remote pathology expertise
4. Resource-poor settings
5. Cancer diagnosis
6. Diagnostic accuracy
7. Patient outcomes
8. Community-based screening
9. Staining requirements
10. Microvasculature analysis
*Insights:*
1. Digital pathology has the potential to revolutionize cancer diagnosis and treatment in resource-poor settings.
2. AI-powered image analysis and remote pathology expertise can improve diagnostic accuracy and reduce turnaround times.
3. Community-based screening and early detection are crucial for improving patient outcomes.
4. Staining requirements and microvasculature analysis are significant challenges in digital pathology.
5. Innovative solutions are needed to address the disparities in cancer diagnosis and treatment in resource-constrained environments.
*Thematic Clusters:*
1. Digital Pathology and AI-Powered Image Analysis
2. Resource-Poor Settings and Community-Based Screening
3. Cancer Diagnosis and Diagnostic Accuracy
4. Patient Outcomes and Innovative Solutions
CC licence: https://commons.m.wikimedia.org/wiki/File:Major_topics_of_pathology_informatics.png#mw-jump-to-license
Prelude:
[07/12, 12:57] SS: If materialized,would be a breakthrough...but the catch is ,would there be explosion of cancers,which otherwise would be indolent.
[07/12, 13:06]rb: Thanks!
Just realised that there was preceding discussion here that I missed adding!👇
[03/12, 21:28] H : Good evening all,
I would like to ask your opinion about the patients coming these days with the AI whole body health screening reports. They are claiming early cancer detection.
Had a patient today with just a single picture of a sub centimetre solitary peripheral lung nodule and nothing else, no x ray, no CT, no radiological report.
Female Patient was completely asymptomatic, with no family history either.
There was also calcium in the coronaries, referred to the cardiologist as well.
How do you deal with these? Are these validated investigations? Are they regulated by the NMC?
[04/12, 07:01] NS: The problem is we don’t have enough data to know if and how these will progress and what to do about them. The idea of screening is to pick up disease at an earlier point in time where the time saved results in improved survival or cure rates. However screening policies at a population level requires it to make financial sense as well. In a country like ours where the health insurance coverage is low and fractured, there is very little cancer screening that gets done and applying a policy is difficult (cf the NHS where the government is the primary insurer). All forms of screening need validation (needs proof for that particular disease) and a clear risk benefit ratio. The concern with whole body scans and other such tools is over investigation with unnecessary biopsies and their complications, anxiety about findings that may not be pathological and financial cost. Hard to regulate in a country like ours where so many different people pay for healthcare in different ways, but professional organisations have guidelines for cancer screening for different populations, and validated tests
[04/12, 07:02] NS: Can’t ban these tests patients should be able to opt for them, but also need to understand it’s not always clear how to interpret the results
[04/12, 07:06]A: This is an evolving area, we do have several ai related publications on different healthcare use cases. They may not be having sensitivity as we want but still it's work in progress. Problem arises when several companies start advertising their product with a bare minimum validation on a small subset. As clinicians, majority of us would look at the evidence before deciding what's best for our patients. But, many of these testing are available directly to patients as well. Their interpretation and utility needs to be taken with a pinch of salt! Medical community as a whole is generally very sceptical in adopting newer things, and rightly so. I have had discussions with a few ai based start up previously where they wanted me to acquire their product. But, I offered to help them provide patients for validation instead.
[04/12, 09:27]rb: Well said and the best way forward for clinicians in the current state of unoptimized clinical complexity veering toward chaos
[04/12, 09:33]rb: Yes and also there's the tumor which could be a turtle or an eagle! Merely detecting the tumor without the ability to tell it's subsequent behaviour other than following up or god forbid killing the turtle for the low hanging fame while getting lacerated by the eagles talons once it begins to perch is any clinician's nightmare!
https://m.northcoastjournal.com/life-outdoors/cancer-part-2-turtles-birds-and-rabbits-12850085
[04/12, 09:38] Group Moderator: fascinating discussion and much needed. What would the clinician community advise AI screening builders to keep in mind to strike the right balances, build credibility and have an impact etc?
No comments:
Post a Comment