Wound infection can occur as a result of injury or surgery. Early detection of infection, and particularly surgical site infection (SSI), is a widespread concern in in both wealthy developed countries as well as in low-resource areas in developing countries. Infections not only increase the cost of healthcare, but can also result in death. Even in the world’s wealthiest countries, approximately 2-5% of hospital surgical patients develop infections, resulting in approximately 0.64% of hospital deaths. In low-resource areas, the incidence of infection is significantly higher, due to many challenges, including a scarcity of medical expertise and a lack of access to water and sanitation. Perhaps more concerning is the increasing prevalence of anti-microbial resistance (AMR) which increases the urgency of detecting and treating infection as early as possible.
Since 2018, our group has been collaborating with Prof Bethany Hedt-Gauthier at Harvard Medical School, and Partners in Health Rwanda, to help mitigate surgical site infection in women who have given birth via Cesarean section. Dr. Rich Fletcher is the technology lead, and our group has been developing the mHealth technology platform (mobile app and central server) as well as the machine learning algorithms to support the work in this research. With the support of multiple grants from NIH, over the past several years, we have worked with the medical team here in Boston as well as the field team in Rwanda to collect data from over 1000 women in Rwanda and develop neural net models that can be run in real-time in the field — either via the server or on the phone itself. (our mobile app uses Tensorflow Lite to enable models to run in Android).
With our current funding from NIH, this technology is being customized and integrated into maternal health services that will hopefully become a standard part of health services in Rwanda. One of the students from our group (Audace Nekeshimana) is from Rwanda and has founded a company in Rwanda (inisghtiv.ai) that was a top award winner at the MIT IDEAS competition in 2019 and is now part of our study team that is helping to commercialize this technology.
While many different technologies were developed to make this study possible, the two main technical innovations introduced by our team are: 1) the real-time computer vision algorithms, implemented on the mobile phone, that enables high-quality image data to be collected from the patients; and 2) Machine learning models (CNN transfer learning) that can be applied to visible images and thermal images to detect infection.
This project has produced three MIT Masters theses and several research publications (both engineering and clinical). (Search on PubMed or Google Scholar) Other articles are in review.
In the videos below, we demonstrate the general Wound Screener mobile app, and the second short video demonstrates the real-time computer vision algorithm. The photo at the top of this page depicts a community health worker in Rwanda taking a photo of the C-section wound of a new mother at her home.
Mobile app developers: Lilian Wang, Rich Redemske
Machine Learning models: Gabriel Schneider (deep learning models), Olasubomi Olubeko (early work), Harsh Sonthalia (early work)
Server platform: John Mofor, Saadiyah Husnoo, Audace Nakeshishimana
Harvard students: Siona Prasad, Joanna Ashby
Rwanda Students: Monique Abimpaye
Collaborators: Bethany Hedt-Gauthier (Harvard), Robert Rivielo (Harvard, BWI), Adeline Boatin (Harvard/MGH), Fredrick Kateera (Rwanda, PIH), Theoneste Nkurunziza (Rwanda, PIH), Vincent Kubaka (PIH), Laban Bikorimana (PIH), Anne Niyigena (PIH)