Author Archive: Soumen Sarkar

AI Doctor is almost ready

Posted on by


For a hospital’s managing director, most crucial and expensive resource is doctors’ bandwidth. Imagine an ailing patient entering the hospital for the first time.  In US, a GP would visit him and run the first-level diagnosis. Based on his analysis a line of treatment would be indicated or the patient would be referred to a specialist.

In India, the patient would go to the the front desk and they will guide to the appropriate department based on patient’s answers.

In both cases, It is humanly impossible to remember about vast number of medical issues with correlated symptoms. This means that initial diagnosis could take a little while, sometimes at the cost of patient’s time and patience.

Now imagine that front-desk is equipped with doctor-grade AI or the GP is assisted with doctor-grade AI, like ChatGPT, trained on vast corpus of medical data and researches. It is somewhat like all the specialists are standing behind the GP to help him reach correct diagnosis at the quickest time possible.
With the help from AI, pin-pointing the right department would be faster for the front-desk assistant. AI can help the GP handle a lot more load with higher accuracy.

Medical Generalist AI

With faster initial diagnosis, hospital’s productivity increases by magnitude. The fact is technology already has reached this level. Commercialization may take another year for the regulation bodies to create guard-rails for the use of such technology ubiquitously. But advanced hospitals may start utilizing the generalist AI service much earlier.

DeepMind Med-PaLM

Particularly two companies are at the forefront now. Google’s DeepMind is almost ready with Med-PaLM. Med-PaLM is a multi-modal large language model (LLM) trained on vast medical data. In the recent report it claims to be 90% accurate in its responses. It’s repertoire is vast if not comprehensive. It can understand Clinical Languages. It can analyze images (radiology etc). It is claimed to speak Genomics too!

OpenEvidence

And there is OpenEvidence, promoted by Mayo Clinic. Theirs’ also is a multi-modal large language model but with an extra chip on its shoulder. Unlike other LLMs like ChatGPT, it is trained with real-time data. With real-time data streaming into it, it would be up-to-date about latest medical findings on every sector –or that’s what they claim when the service will be commercially ready. It could be very soon since they recently claimed[2] it to score above 90% in United States Medical Licensing Examination.

OpenEvidence AI claims to become the First AI in History to Score Above 90% on the United States Medical Licensing Examination (USMLE)

Opportunity or Threat?

GPOnline finds that England needed nearly 7,400 more full-time GPs to manage the patient care. According to a 2021 WHO report, India was just above the 2006 standard of 22.8 healthcare workers per 10,000 population. With professional doctor-grade generalist AI at disposal of every hospital, relief is not going to be quantitative alone, it can add qualitative improvements in patient diagnosis in terms of reducing diagnostic time, improving accuracy and potentially reducing number of tests for the patient.

With a professional team always monitoring the model’s answers and fine-tuning the model’s learning, the improvement will be continuous and so will be the efficacy of the system.
Following the current trend, entire service most likely will be made available as SaaS for the hospitals and clinics. SaaS model also will help bringing incremental cost of adoption down to an affordable basis.

In summary, it looks like that generalist medical AI is going to bring paradigm-shift in patient care in not a very distant future. How pervasive the adoption would be though would depend on a hospital’s patient-care economics.

Read more

1.DeepMind announces Med-PaLM
2.First AI in History to Score Above 90%
3.OpenEvidence official site
4. Med-PaLM official page

Understanding asset tag technologies

Posted on by

Indoor asset tracking is coming to mainstream now. Whether in hospital, or in warehouse, people are realizing that there are certain benefits that they cannot ignore any more in their business context.

But before you adopt it, would you not like to understand what it takes to implement? Of course there is this software called Real Time Location System (RTLS) but what about the tags? This is an attempt to make it simpler for you to decide.

Many Tag technologies

Every asset tracking system requires tags to be fixed on your assets. Now there are different tag technologies, like RFID, LoRA, BLE, UWB etc.

RFID can be both passive and active. Passive RFID tags do not have battery and therefore do not transmit by itself. RFID readers in vicinity sends search packets and the passive tags respond to them (they are passive after all!). Active Tags regularly send data which nearby Reader/Gateway captures. Sometimes people use Active RFID interchangeably with BLE or Bluetooth Low Energy.

So which tag should you choose?

The plain answer is it depends. For example, not all RTLS vendors support all tag types. In most cases vendors sell tags and other hardware as part of the full package solution. Other important factors depend on what you actually need.
Passive RFID tags cost the least among all of them but RFID readers some times miss reading passive RFID tags. In fact reading consistency is about 90%-95% for RFID readers with UHF antenna (UHF is the fastest among all the RFID system).

On the other hand UHF RFID readers are very expensive. Compared to that BLE devices cost magnitude cheaper. BLE is more accurate as well. UWB is the newest of the technologies and the technology is capable of precisely location an asset within a few centimeters. Hospitals traditionally went with RFID tags, many with vendor-proprietary active RFID solution. Proprietary solution makes you locked-in with vendor.

Incidentally INDTRAC RTLS works with almost all different hardware and tags and lets you change your tag system in future. For example, let’s say you decided to adopt BLE now. Later if you decide to upgrade to UWB, you can do that with INDTRAC.

Now coming back to relative benefits and costs. Typically UWB is the costliest solution at this point but offers most precise location articulation.

Here is a simple comparison chart for all the tags for quick reckoning.

Feature RFID BLE LoRa UWB
Technology Radio Frequency Identification Bluetooth Low Energy Long Range Ultra Wide Band
Range Short (~ 20 feet) Short (~ 40 feet) Very Long (~ 1 mile) Long (~ 100 meters)
Accuracy Low (~ 10 feet) Medium (~ 2 feet) Low (~ 100 feet) High (~ 1 inch)
Power Consumption Low Low Low Medium
Tag Cost $0.10 – $1.00 $0.50 – $5.00 $1.00 – $10.00 $5.00 – $50.00

Summary

In summary, if you follow traditional solutions, you would use RFID. If accuracy is paramount for you, you would have to choose UWB. But in most cases, BLE serves both ends. It also offers relatively long battery life for tags. Gateways also do not cost as much as UHF RFID. They are also easier to install or replace. INDTRAC supports both Indian brands as well as international brands like Zebra , Impinj or INGICS .