AI in Healthcare is More than an AI Physician
Autonomous driving, to put it simple, is about training an AI driver. Of course there are a lot of things and technologies behind that but the idea is to remove people from the driver seats when transporting people from point A to point B.
“Autonomous healthcare”, on the other hand, is way more about training an AI physician, as naturally, it takes the more than one physician to keep people stay healthy. And the interpersonal relationship between patients and doctors/nurses are much more intense than that between passengers and drivers.
As complicated as it could be, healthcare is one of the thriving vertical to leverage AI, as traditionally, most of the medicare decision has been driven by data and rules (from generations of doctors’ experiences and researches), regardless the data is the reading from CT or blood test, or by “observation, listening, interrogation, and pulse palpation” by an experienced Chinese medicine doctor. When the computer age came, a lot of data has been digitalised with more and more testing and scanning technologies been developed and deployed. More data from easily-available IoT sensors and computing power makes below few hot areas for AI in healthcare:
- Medical image reading and supportive diagnosis: it’s no longer a news that AI can identify picture of cupcake-like-dog from pictures of real cupcakes, so it’s natural that AI could also read broken arm (radiology), count cell (microscopy), identify cancer cell (digital pathology) or evaluate skin condition. Similar to the regulation case in autonomous driving application, we won’t see an fully autonomous AI doctor anytime soon, but AI would help save doctors’ time during the diagnosis process.
- Motion analysis: similar to application in medical image, AI can help physician to read video. Especially in the case of rehabilitation when patients try to regain the capability of daily living, a depth camera could help the physician read accurate motion and progress of the patient so as to make adequate instruction, even remotely. Patients on fertility, for example, would also get help from AI to improve the success rate. The other application would be for day care service; AI can help the service providers manage the facility, detect or predict potential risk like fall down, etc.
- Robotic: by integrating robotic and AI in both image and motion analysis, an AI-assisted surgery has been underway. Some startups also integrate AR/MR for surgeon to overlay data and suggestion from the cloud when practicing with robotic surgery.
- Personalized drug/treatment: a comparative trend in the cosmetic industry is about personalization. Based on analysis on the living habit, skin condition, environment, etc. a cosmetic company could customize the product for consumers. Similar thing on healthcare, just longer way to go. There are data from personal wearable device like Apple Watch, there are health record if it’s digitized and shared through a well-design healthcare system. Marrying AI with genetic also enable the possibility of personalized drug in longer term.
- Hospital IT and management system: application ranging from more traditional dosage error prevention, fraud detection to hospital facility management
In short, when we talk about AI in healthcare, it’s about building an AI-backed healthcare system, from health management, “pre-medical”, supportive medical IT, devices and equipment, hospital management, all the way to post-medical supportive systems. For that, data with quality plays even more critical roles than other sectors. We may see Waymo cars test-ride in California with a human driver remain in the seat, it’s not easy to see comparable model in AI healthcare. Data like medical image, unlike road, is typically held within the hospital or clinic and not open for access unless there is an approved research project under public supervision.
To make the algorithm prediction work, repetitive (meaning, medical record from (ideal) same patient over a certain period of time), multi-dimensional (readings and annotation from different clinical sector) data is a key. So a place that patients go to see doctor quite often and the medical record could be digitalised and annotated clinically would be ideal. Place like Taiwan with well-developed healthcare system, as many hospital and clinics combined as convenient stores, one of the highest doctor-seeing per year per capita, would be an ideal choice to look for clinical or academic partner. …shared by Yu-Chuan (Jack) Li, Dean, College of Medical Science and Technology, Taipei Medical University, during Demo Day#2 of BE Accelerator
Ultimately there is a need for system integration to put together all the puzzle, from pre-medical like genetic analysis and environmental data collection to detect and predict high health risk factor to healthcare, new way of health data collection like blood test at home, to more effective and efficient medical practice to post-medical service. There is much more to be done and there is definitely a great opportunity for corporate and startup to work together and innovate.
For reference: here in Taiwan Tech Arena I’ve been meeting with a few amazing founders on this topic.
Medical image: aether AI dedicates to developing intuitive web interface for medical images and AI-powered image analysis workflow.
Motion analysis: Longgood developes rehabilitation training programs and physical fitness assessment tools
IT: Ucare.ai builds an AI-powered predictive hospital bill estimation system
Robotic: Brain Navi is a AI-powered surgical navigation system for brain surgery
Also quite a few more from TTA’s accelerator partner BE Accelerator and check out their latest Demo Day.
