Artificial Intelligence (AI) is indeed the transformational technology of our generation and whilst its practicality in Healthcare is still yet to be fully felt – real-life deployment and funding of promising AI startups are ever-growing. We predict that within the decade, Artificial Intelligence would be recognized as a complementary arm of healthcare – and while it might not fully replace the traditional role of professionals, AI certainly can disrupt healthcare to become better, faster, cheaper and more accessible.
The benefits of Medical AI in unprivileged communities cannot be overstated. AI could save lives (and your pockets) on a global scale.
So in this emerging evolution from codes to clinic, which are the movers and shakers in this promising industry? We examine a few as they will, de facto, shape our future.
Stanford Centre of Artificial Intelligence in Medicine & Imaging (AIMI)
United States of America
[Status] R&D-in-progress, Private University
What they are working on: Radiology Deep Learning, Data Augmentation, X-ray/plain film analysis, volumetric film analysis
Why they are promising: Stanford is a world renowned higher learning institution with her own medical facilities. Comprising a famous faculty including the likes of Prof Curtis Langlotz, Stanford also has immense funding, manpower to fund their cutting edge research. Particularly impressive is the fact that Stanford organizes her own Medical Imagenet though the dataset is not yet made available to the public.
Potential challenges: Deep Learning for radiology is perhaps not the easiest domain to tackle. The very fact that many radiology models require a volumetric assessment (eg: CT/MRI) and may not be clinically significant if interpreted by frame means that Stanford AIMI has quite a battle. Labeling in radiology is also another pandora box. Consolidation or frank pneumonia? To tackle this, Stanford AIMI will probably need to standardize and protocolize these labelings.
Medical Good Intelligence Company (MEDGIC)
East Coast Road
[Status] Active, Non-Profit, Private Startup (Independent)
What they are working on: Dermatology Deep Learning, Bayesian Network, Skin analysis, Skin Diagnosis, Skin Scanning, Detection of skin conditions
Why they are promising: Probably the first and only on earth to tackle the entire spectrum of dermatological conditions (with pretty impressive accuracy for a start), Medgic is a non-profit startup which has resisted funding and to date remains strictly independent. It is made in Singapore by a team of doctors and computer scientists. The founder, Dr R Lim has stated that funding from private funds “would make Medgic chase money” but “the real ROI is the number of lives we could change.”
While Medgic is not rated as a medical device nor does it claim to be diagnostic or have any medical accuracy yet (pending clinical trials which would require larger funding), it is released as a general skin scanner, detector and analyzer to “provide general information for general purposes only”.
Potential challenges: Being independent and small, the team will have a harder time marketing itself and making the technology known to the remote corners of the world. Language barriers is probably the next big challenge – Medgic AI actually asks users about their skin condition – and medical terminologies often are hard to translate. This is especially true as Medgic is trying to target unprivileged communities with poor access to healthcare.
Mountain View, California
United States of America
[Status] R&D-in-progress, Private Corporation
What they are working on: Retinography Deep Learning, Pathology Deep Learning
Why they are promising: Google AI is no doubt the grandfather of Artificial Intelligence innovations. The world stood in awe as AlphaGo beat Kai Jie in the ancient Chinese game of Go back in 2017. Since then the team has been working on meaningful applications such as medical imaging (as above). The sky is indeed the limit for Google.
Potential challenges: Pathology is an important area to disrupt but this field is relatively operator dependent. For example, pathology slides may differ significantly depending on the operator’s choice of, amount and combination of staining. Retinography similarly requires the use of a complex machine which many professionals might still not be skilled at. That being said, these are surmountable challenges which can be tackled over time.