AI as diagnostician in ophthalmology, dermatology. Faster adoption than IBM Watson?

Three recent articles from the IEEE (formally the Institute of Electronics and Electrical Engineers) Spectrum journal are significant in pointing to advances in artificial intelligence (AI) for specific medical conditions–and which may go into use faster and more cheaply than the massive machine learning/decision support program represented by IBM Watson Health.

A Chinese team developed CC-Cruiser to diagnose congenital cataracts, which affect children and cause irreversible blindness. The program developed algorithms that used a relatively narrow database of 410 images of congenital cataracts and 476 images of normal eyes. The CC-Cruiser team from Sun Yat-Sen and Xidian Universities developed algorithms to diagnose the existence of cataracts, predict the severity of the disease, and suggest treatment decisions. The program was subjected to five tests, with most of the critical ones over 90 percent accuracy versus doctor consults. There, according to researcher and ophthalmologist Haotian Lin, is the ‘rub’–that even with more information, he cannot project the system going to 100 percent accuracy. The other factor is the human one–face to face interaction. He strongly suggests that the CC-Cruiser system is a tool to complement and confirm doctor judgment, and could be used in non-specialized medical centers to diagnose and refer patients. Ophthalmologists vs. AI: It’s a Tie (Hat tip to former TTA Ireland Editor Toni Bunting)

In the diagnosis of skin cancers, a Stanford University team used GoogleNet Inception v3 to build a deep learning algorithm. This used a huge database of 130,000 lesion images from more than 2000 diseases. Inception was successful in performing on par with 21 board-certified dermatologists in differentiating certain skin lesions, for instance, keratinocyte carcinomas from benign seborrheic keratoses. The major limitations here are the human doctor’s ability to touch and feel the skin, which is key to diagnosis, and adding the context of the patient’s history. Even with this, Inception and similar systems could help to triage patients to a doctor faster. Computer Diagnoses Skin Cancers

Contrasting this with IEEE’s writeup on the slow development of IBM Watson Health’s systems, each having to be individually developed, continually refined, using massive datasets, best summarized in Dr Robert Wachter’s remark, “But in terms of a transformative technology that is changing the world, I don’t think anyone would say Watson is doing that today.” The ‘Watson May See You Someday’ article may be from mid-2015, but it’s only this week that Watson for Oncology has announced its first implementation in a regional medical center based in Jupiter, Florida. Watson for Oncology collaborates with Memorial Sloan-Kettering in NYC (MSK) (and was tested in other major academic centers). Currently it is limited to breast, lung, colorectal, cervical, ovarian and gastric cancers, with nine additional cancer types to be added this year. Mobihealthnews

What may change the world of medicine could be AI systems using smaller, specific datasets, with Watson Health for the big and complex diagnoses needing features like natural-language processing.

Lūbax skin cancer detection app in clinical trials

What is that spot? A question that many of us have worried over. A skin cancer detection app developed by Los Angeles-based Lūbax is being tested with physicians and dermatologists in the US, Australia and the UK as of 4 February, World Cancer Day. It uses image recognition software and algorithms to search a proprietary dermatology image database of over 12,000 diagnosed lesions. Their initial large melanoma clinical trials with Harvard, Stanford, Oxford, and the University of São Paulo showed sensitivity and specificity in detecting large melanomas in patients. According to their website, they are expanding the database and algorithm to include basal and squamous cell carcinomas in addition to amelanotic melanomas. There’s a signup for the clinical trial only for iPhone, but their execs in a news item have met with Samsung. BioSpectrum Asia, release on, startup profile on

Medical informatics in dermatology course (NY)

Thursday 13 November (all day), Memorial Sloan Kettering Cancer Center, Rockefeller Research Laboratories Auditorium, New York City

‘Envisioning the Future of Dermatology Through the Lens of Medical Informatics’  will enable the clinician to better understand:

  • the novel approaches to diagnose skin cancers with computer informatics
  • the benefits, limitations and integration of photography and electronic medical record in dermatology
  • the benefits, limitations and legal barriers in teledermatology
  • the factors related to privacy issues for images

6.0 AMA PRA Category 1 Credits. Early registration savings before 1 October: $50 MDs, PhDs and DOs; $25 Fellows, Residents, and RNs. After 1 October add $25. More information and registration. Hat tip to Peter Brodhead of MSKCC and Howard Reis of HealthePractices.