Operations and Implementation
EP-138 - Implementation of a Public Tele-Dermatology Program in Brazil and Development of Image Deep Learning Algorithms
Monday, April 30
9:35 AM - 9:50 AM
Location: Education Zone, Booth 2416, Screen 4
Skin diseases include more than 2000 specific conditions, from simple rashes to malignant lesions and life-threatening anaphylactic responses. Although most rashes are easily managed by a well-trained general practitioner or internist, patients are often referred to consultation with a dermatologist. In those cases, long waiting times may worsen disease outcomes and increase health overall costs.
São Paulo is the largest and most populous city in Brazil. Although Brazilian public Unified Health System guarantees universal health coverage, there is great inequality of access to good quality specialized healthcare, mainly in low-income areas. By July 2017 there were 477 public primary care facilities working in São Paulo, with a mean 16000 patients referred monthly to dermatologist consultation, exceeding the public system capacity of 10000 visits per month. This imbalance lead to a mean waiting time of six months and a 65000+ patients long queue.
Since dermatologic examination is mostly visual, the field of tele-dermatology has grown exponentially with the recent advance of high definition image communication technologies. Tele-dermatology programs are already consolidated in well-developed countries and accuracy has been widely accepted to be comparable to in-person visits.
In early 2017, the city Health Department partnered with Hospital Israelita Albert Einstein, one of the country's most important private non-profit healthcare organizations, to develop a tele-dermatology program, aiming to reduce waiting times, increase quality of dermatologic care, reduce costs, and collect data to develop an artificial intelligence algorithm able to assist on skin lesions diagnosis, a full triple-aim program.
In brief, a web-based platform was developed in-house to allow secure storage of images and patient information, available to trained hospital staff dermatologists for adequate documentation of diagnosis and recommendations. Trained nurses are responsible for filling patient data forms and perform the collection of digital images for asynchronous image and clinical data evaluation by a staff dermatologist. Diagnostic hypotheses, treatment options and recommendations are recorded for each case. Eventually the patient receives one of three evaluation results: 1) Definite diagnosis and treatment recommendation - return to primary doctor; 2) Suspected malignant lesion - ordered skin biopsy; or 3) Unable to give a definite diagnosis - referral to in-person dermatologic consultation.
Along with the patient-centered strategy, a dedicated technology staff was designed to develop an artificial intelligence algorithm using acquired image data. Deep learning techniques are being used to automatically suggest a probable diagnosis for skin lesions.
This e-poster will present the tele-dermatology program development, its implementation challenges and preliminary results. Over the first 40 days of implementation, 2054 skin lesions were evaluated. 47.9% of the patients were given a definite diagnosis and avoided an in-person consultation, 46.4% were referred to traditional care and 5.7% were directly referred to skin biopsy. The early-scale project is designed to evaluate 65000 skin lesions by early-2018. Deep learning algorithms are being generated and details about the developing process will be presented, as well as its first accuracy results.
- Define main steps to implement a tele-dermatology service able to reach thousands of patients
- Determine the cost-reduction potential of a tele-dermatology approach
- Get acquainted to artificial intelligence algorithms being developed to automatically screen skin lesions