January 30, 2020
By Charles C Onu (AI Research Lead)
Ubenwa's journey of building clinically-backed, low-cost diagnostic tools for newborns is a long one! Last year we received ethics board approval to launch our clinical trial in Nigeria and Canada. We are now gathering the largest database of clinically-annotated infant cries with real patients. As we begin the new year, we would like to share some of the major milestones of 2019 and our plans for 2020.
Ubenwa is a mobile app that analyses the cry sounds of a newborn to detect early signs of perinatal asphyxia - a leading cause of neonatal disability and death. The app uses machine learning to identify the changing acoustic patterns in the cries of newborns who are at risk of brain damage due to asphyxia. It alerts care-givers of infants at risk, allowing them to apply necessary treatment and/or make an early referral to tertiary care facilities. See our short video below to learn more.
In 2019, my team had one main target. We wanted to launch clinical studies to validate Ubenwa within hospitals. By working methodically with the team and our advisors, we overcame several obstacles in launching our clinical trials. We first sought out and recruited new clinical partners. Then in collaboration with our partners, we designed and developed detailed protocols for the clinical studies. We applied for, negotiated and eventually obtained ethics board approval, both in Canada and Nigeria. With grant funding, we developed a robust technology platform for the clinical project.
We finally hit the critical milestone of launching the clinical study in the last quarter of 2019, after over 18 months of work. The study was kicked-off first in Nigeria, at the Enugu State University Teaching Hospital (ESUTH), and to begin subsequently in Canada, at the Montreal Children’s Hospital (MCH). We are now gathering the largest database of clinically-annotated infant cries while also testing Ubenwa with real patients. Our goal is to acquire up to 10,000 cries from 2,500 patients within 2 years. This will ensure that we have the right amount and diversity of data to demonstrate the efficacy of Ubenwa.
As expected at the start of projects of such scale, we have already had to deal with a couple of new, unforeseen challenges. These have included both technical design and operational issues regarding the clinical workflow. We continue to put our heads together to address these swiftly. In October 2019, Samantha, clinical lead of Ubenwa, spoke about the clinical studies and issues surrounding them in Munich at the annual UBX19 Digital Opportunities conference. See video below:
In March 2019, we were honored to receive the World Health Organisation (WHO) award as one of the top 30 health innovations in Africa. Innocent (Software Engineering Lead) represented us in Cape Verde at the 2019 Africa Health Forum.
Our scientific paper was accepted and presented by me at the Annual Conference of the International Speech Communication Association ( INTERSPEECH 2019). With about 3,000 attendees, INTERSPEECH is one of the largest venues for researchers doing speech and machine learning work. This paper was one that I was quite excited to talk about because we made an interesting finding. Our investigation revealed that representations learned from adult speech could actually inform and improve performance of models developed for infant cry analysis. We also found that such models were more resilient to different types and degrees of noise. This paper was titled “Neural Transfer Learning for Cry-based Diagnosis of Perinatal Asphyxia”. You can read it in full on Arxiv.
Throughout the year, we continued to engage, interact with and learn from the broader community involved in machine learning and global health. Some of these programs are:
“What physicians want to see is the data. Show them that you have conducted clinical trials, that you have tested on thousands of actual patients, and that the app demonstrated good sensitivity and specificity” - Dr Datonye Briggs, RSUTH
A week ago, I attended the annual conference of the Paediatric Association of Nigeria. My goal was to engage closely with the local paediatrics community and to interview practitioners to develop a map of how perinatal asphyxia is managed around the country. At the end of each interview, I asked the doctors one question: what will it take for you to accept and begin to use a tool like Ubenwa in practice. Their answers were always akin to that of Dr Briggs above.
Finally, our work is rapidly beginning to outpace our current team. We will be actively looking for ways to bring on new team members to meet the growing data collection requirements as well as to pursue new research questions using the unique data that we are acquiring.
We are grateful to all our advisors - Prof. Doina Precup, Prof. Yoshua Bengio, Prof. Edward Alikor, Prof. Robert Kearney, Scott Schaffter, Dr Gautam Bhattacharya, Dr Eyenimi Ndiomu and Urbain Kengni - for their steadfast belief in us. We thank Mila, District 3, Jeanne Sauve Foundation, Ministère d’Economie et d’Innovation, and Creative Destruction Lab for their support. I personally thank Vanier CGS for supporting my own research and our clinical collaborators - Prof. Guilherme Sant’Anna (MUHC) and Dr Uchenna Ekwochi (ESUTH) - who have believed in our vision and are joining us to make it a reality. And to everyone who continues to support our work in any small way, we say a big thank you.
Happy new year from all of us at Ubenwa.