More than 700,000 Rohingya refugees have fled Myanmar in the past year, finding temporary shelter across the border in Bangladesh’s Cox’s Bazar district.
Their overcrowded, unhygienic living conditions make them especially vulnerable to infections and other diseases — a need the United Nations has called “immense,” with healthcare services “grossly under-funded.”
Help in addressing these distant needs is coming from a Silicon Valley healthcare startup, Polyfins, whose AI-powered mobile app can help dermatologists diagnose skin conditions that often hold strong clues to underlying disease — ranging from common conditions like eczema to serious ones like tumors, infections and parasitic infestations.
“In a refugee camp, there are a lot of people staying in a very small area with very unhygienic conditions,” said Faisal Basar, Polyfins’ CEO. “This is fertile ground for contagious diseases with skin manifestations.”
Polyfins, a member of the NVIDIA Inception program, worked with Bangladeshi authorities to bring two dermatologists and several nurses to the refugee camp. The team saw around 500 patients over two days.
The dermatologists used Polyfins’ Tibot smartphone app as a diagnostic aid on one day, and used traditional diagnostic methods on the other. They found that they were able to see 25 percent more cases if the patients’ symptoms were first analyzed by a nurse using the app.
AI as Diagnostic Aid
Tibot requires users to upload an image of the skin condition and answer questions about their symptoms. The responses and a compressed version of the image are then sent to Polyfins’ server, which sends back its prediction of the top three skin conditions the user might have.
The app’s analyses allowed the doctors to see more patients at the medical camp, which in turn enabled Polyfins to distribute much-needed medication to a larger number of patients in the camp.
The AI behind the app was trained on a dataset of 100,000 images from public sources, research partners and hospitals Polyfins works with. While developing its app, the startup collaborated with doctors to label the image data and provide guidance on identifying disease symptoms.
Polyfins uses NVIDIA GPUs in Google Cloud for training and inference of its neural networks, which were developed using the TensorFlow framework.
Image compression allows Tibot to efficiently run its analysis in the cloud — even in areas with 2G or 3G cellular networks, like the refugee camp. Since launching in June, the app has analyzed 15,000 cases.
Beyond Diagnosis
Based in Fremont, Calif., with teams in India and Bangladesh, Polyfins is also partnering with dermatologists at the D. Y. Patil Hospital in Mumbai, India, on a forthcoming paper to evaluate the accuracy, sensitivity and specificity of their deep learning model.
Basar sees the tool as a more effective alternative for people who use a search engine to figure out what conditions they might have.
The team plans to add additional features to the app beyond diagnostics — such as care recommendations, virtual visits with physicians and follow up of skin lesions over time. Polyfins’ goal is to make Tibot a complete skin management app that leverages AI.