Agenda Items:
-
How to size an archive for a hospital? (# xrays, cts, etc)
- http://www.pacscalculator.com/
- Market Definition
- Hospitals/Health Systems
- T Shirt Sizes : S, M, L, XL (20+ in US - Kaiser, MGB, HCA, CHI)
- Small - 0-50,000 studies/year -
- Medium - 50,000 - 200,000 studies/year
- Large - 200,000 - 1.5M /uyear - 80 rads / 1M Studies
- XLarge - 1.5M+
- Imaging Centers
- Rad Groups
- Teleradiology
- Hospitals/Health Systems
- Which departments?
- General Radiology -
- 40% X-Ray
- 70% 2 View Chest Xrays - 24MB UN/10MB C
- 10% 1 View Portal XRays - 10MB UN/5MB C
- 20% - other stuff
- 20% CT
- 300-2000 images - average 500
- 10% MRI
- 300-18,000 images - average 400
- 10% US
- ??
- Everything else - 20%
- 40% X-Ray
- General Radiology -
- Most hospitals know there data (modality mix, # procedures, # rads, type of scanner)
- One common view is that average study size is 50MB (skewed by high number of XRays)
-
Protocol Versioning
-
PACS Pricing Models
- Pay per study
- Pay per consumption (like cloud)
- How are old exams going to be paid for?
-
Initiatives that the industry(we?) should pursue in 2022
-
Predictions for industry on Dec 31, 2022
-
Trends
- Home/Virtual healthcare
- Modality as a service?
- Point of care ultrasound (butterfly)
- Hyperfine / Portable MRI
-
Technology that other verticals are leveraging that should be coming into the industry
- AI/ML
- Value based care - how it relates to AI/ML
- Metaverse?
- Blockchain?
- Agile vs Waterfall BUSINESS MINDSET
- Cloud (duh - no brainer)
- Object Storage
- Serverless technologies
- Managed Databases
- Infrastructure as code (IAC)
- Tyto Care Device
-
Current situation in american hospitals (covid)
-
Brad Genereaux to Everyone (10:04 AM) This one I think you mean - http://www.pacscalculator.com/ STEVEN BORG to Everyone (10:05 AM) Yep, that's the one! Sandor dr Konya to Everyone (10:18 AM) cardiology fluoroscopy / qngio suite images can be incredibly huge. Simon Rascovsky to Everyone (10:19 AM) [file: Screen Shot 2022-01-03 at 11.18.53 AM.png] That’s just a volume distribution for a large academic site. Not storage size. Brad Genereaux to Everyone (10:20 AM) One of my favorite articles looking at business models in medical imaging; focuses on AI but the concepts can be somewhat generally applied - https://www.signifyresearch.net/medical-imaging/business-models-ai-medical-imaging/ Sandor dr Konya to Everyone (10:20 AM)
That’s just a volume distribution for a large academic site. Not storage size. ok STEVEN BORG to Everyone (10:24 AM) Thanks! Simon Rascovsky to Everyone (10:26 AM) There’s a lot of variability in reading volumes per rad and modality. this is a year’s worth for an institution with ~500,000/year, 150+ rads: Simon Rascovsky to Everyone (10:26 AM) [file: Reading provider volume per modality.png] STEVEN BORG to Everyone (10:28 AM) @Jean-Francois, you make an EXCELLENT point. I've considered it, but was looking at 3x retrieval per first year (then down to 1/2 year), not 7x. Is 7x per first year? Jean-Francois Pambrun to Everyone (10:30 AM) This ~7x figure is from historical data. Counting coordinator, QA, residents, viewing priors etc. I don't know about the average timelines. STEVEN BORG to Everyone (10:32 AM) Thank you! Very valuable @simon, what is the X-axis on your graph. Jean-Francois Pambrun to Everyone (10:36 AM) pay per view means shifting a big part of the risk on the consumer.. a likely though sell.. Simon Rascovsky to Everyone (10:37 AM) @steve: It’s individual radiologists at that org. STEVEN BORG to Everyone (10:37 AM) Got it! Thanks! Sandor dr Konya to Everyone (10:40 AM) how would industry solve research purpose study retrievals? 🤔 if I would like to access 100k xrays for a classification problem, it would cost a fortune from cloud wouldn't it? Daniel Mwambi to Everyone (10:41 AM) PACS-coin 😅 Danyell Wilt to Everyone (10:42 AM) Sd - Do the research in the cloud rather than egressing the data? You can do the classification using resources in the cloud without needing to move the dataset. Jean-Francois Pambrun to Everyone (10:42 AM) do your ml in the same cloud so you don't have to pay egress Sandor dr Konya to Everyone (10:43 AM) oh, so I pay then for computation too... beautiful future for self initiated research =/ Jean-Francois Pambrun to Everyone (10:46 AM) you need money to make real research on pixel data.. v100 GPUs or whatever are few $ per hours and you can spin up/down in few seconds.. Jean-Francois Pambrun to Everyone (10:47 AM) engineer cost 70-100+$/hr, you should be able to pay few dollars per hours for GPUs STEVEN BORG to Everyone (10:47 AM) https://www.tytocare.com/ https://www.veri.co/ Daniel Mwambi to Everyone (10:51 AM) Could point of care ultrasound replace the stethoscope in the future? Danyell Wilt to Everyone (10:51 AM) Locums work Sandor dr Konya to Everyone (10:52 AM) the problem is not the few bucks / hours... rather the time. now I can buy a few rtx30XX for some thousand $ that run 7/24 for few years and can get the job done... run a cloud computation (amd re run 100x) ... noone can afford it in the long run. Simon Rascovsky to Everyone (10:53 AM) @Daniel, the stethoscope, it could but it’s also much more complex to use, more expensive and there are hundreds of years of common knowledge on chest sounds that is useful. https://cuehealth.com https://detect.com Jean-Francois Pambrun to Everyone (10:54 AM) 8gb of VRAM is limiting for one. Seconddly, you could wait a year with 4000$ of consumer GPU or spend a few grand over night and get the result the next morning.. Sandor dr Konya to Everyone (10:55 AM) sure, I agree, thats an option and if the research has unlimited funds is the way to go =) Simon Rascovsky to Everyone (10:56 AM) https://youtu.be/tEt51zu5gm4?t=21 Jean-Francois Pambrun to Everyone (10:56 AM) you (and academia) consider your time not to be worth enough IMO Sandor dr Konya to Everyone (10:57 AM) @Jean =)=)