In his final year as a medical student, Tobi persuaded the university to create a learning management system for his boring anatomy lectures. He is now using artificial intelligence to help doctors in Africa process medical records faster.
There is a shortage of health workers in Africa. As such, healthcare workers are expected to see more patients per doctor than in developed countries. Paperwork occupies a significant amount of time at work. Healthcare professionals must take patient histories, fill out forms, and update these records from time to time. Modern hospitals are digitizing the way these records are obtained, but computers and some software don’t always make the job easier. Sometimes it takes so long that the doctor goes back to writing notes on paper.
A Nigerian-trained doctor and AI expert believes artificial intelligence can help hospitals in Africa digitize medical records faster, saving doctors time.
Africa’s most populous country, Nigeria, needs 363,000 doctors to achieve universal health care. However, Nigeria has only 24,000 licensed doctors. In 2017, when Nigeria needed 237,000 doctors to meet the World Health Organization (WHO) recommended doctor-to-patient ratio, the country had 35,000 more doctors. Despite an average population increase of 2.46% from 2017 to 2021, the number of qualified doctors fell by 31.4%.
The result is too few doctors seeing too many patients. Nigerian doctors frequently complain of burnout and being “overworked and underpaid”. The story is the same for much of Africa. There are about 3.6 million health workers in 47 countries, and the ratio of health workers (including doctors, nurses and midwives) in Africa is 1.55 per 1,000 population, he said, WHO estimates. doing. The recommended threshold is 4.45 HCPs per 1,000 patients. By 2018, only four countries (Mauritius, Namibia, Seychelles and South Africa) had exceeded the WHO health worker to population ratio.
Digitization of healthcare
In the early 1960s, the Mayo Clinic in Rochester, Minnesota, USA, was one of the first major medical centers to install an electronic health record (EHR) system. It was expensive, basic, and could only be used to manage patient appointments and billing. Since then, electronic medical record software has become much more sophisticated, allowing it to collect and process detailed information about a patient’s health.
EHRs are not widely used in African healthcare settings, despite their much-discussed benefits. Installing an EHR system and training healthcare workers to use it is expensive. Also, the hospital is often criticized for not implementing his EHR tool. But there are more subtle reasons. Tobi Oratunzi, a doctor-turned-computer scientist trained at the University of Ibadan, says that even in hospitals where EHR systems are in place, doctors find EHR systems cumbersome and time-consuming. Says it’s not always used. Doctors may be computer savvy, but “putting a keyboard in front of a person presents an entirely different problem.”
So Oratunzi co-founded Intron Health, a startup that uses automatic speech recognition (ASR) technology to transcribe doctors’ speaking notes. But Intron Health didn’t start with speech-to-text software. Founded in 2019, Intron Health provided regular his EHR software solutions to help hospitals digitize processes.
In 2020, as healthcare workers fear that the COVID-19 pandemic will easily overwhelm Africa’s fragile healthcare system, Intron Health of hospitals piloted the first software. “The hospital was very crowded and everyone was excited. They had electricity. We had a wireless network set up for them. . “[But] The day we started the service, doctors spent 40 minutes just typing notes for their first patient. “It took her 50 minutes to see the next patient, by which time the patients in the waiting room were visibly frustrated. When a doctor is unable to see a patient because creating or updating medical records requires the use of cumbersome computer software, waiting patients may be tempted to seek help elsewhere. A hospital using Intron Health’s early software asked if the software could be simplified by replacing text boxes with checkboxes. But that was a crude solution, and meant that to create a robust enough checkbox system, you would have to anticipate every conceivable medical situation. It was impossible.
When Olatunzi was a medical student at the University Hospital of Ibadan (UCH), during an anatomy class, the instructor struggled to explain how a baby travels through the birth canal using only textbook pictures and hand gestures. and faced similar frustration. He felt that video lessons were better and students could repeat the lessons as often as needed. Somehow he managed to convince the university to build a rudimentary learning management system with hosted animated video lessons, with funding from the National Institutes of Health (NIH) and the World Bank. I was. Since then, Oratunji’s path began to diverge from medical practice. His university asked him to help train staff from other universities, and upon graduation, he was hired by the university to build technology tools such as telesurgery tools, patient navigation apps, and clinical simulation software. rice field.
Mr. Olatunzi came to the United States from the University of Ibadan and received a master’s degree in medical informatics from the University of San Francisco and a master’s degree in computer science from the Georgia Institute of Technology. He was hired as a machine learning scientist and researcher at his Bay Area company Enlitic in San Francisco, where he developed natural language processing (NLP) and natural language understanding (NLU) AI models for translating English text into other languages. helped build the After leaving Enlitic, he joined his Health AI team at Amazon Web Services as a machine learning scientist. Meanwhile, he has already built software to digitize hospital records in Africa, which became Intron Health. However, his first rodeo ran into an obstacle. Doctors in Africa saw too many patients to populate fast enough for a digital medical record system to make his EHR implementation worthwhile.
keyboard to mic
Taking inspiration from the speech-to-text software used by Oratunji’s wife, who is also a doctor, Intron Health tested popular speech-to-text platforms and found the obvious. They all failed to properly transcribe their African accent pronunciations and names. “I tried it myself and found that they weren’t tailored to our accent. explains.
But he was working on this kind of problem, so Intron’s team built a proprietary speech recognition platform and embedded it into the EHR platform to make note-taking easier, freeing up valuable time for doctors. I decided to.
This is 2021 and the Intron community has gone all out to procure African accent speech datasets. Intron currently builds over 11 million voice samples of her and over 200 unique accents from her 7,000 speakers in 13 countries. This makes it possible to build automatic speech recognition software that can translate accented speech with greater fidelity than Google Assistant, Siri, Alexa, etc.
Intron asks healthcare workers to read a short 74-word sentence when enrolling hospitals on its platform. “Most people take up to 55 seconds to dictate this, but the rate of speaking is fairly constant,” Olatunzi explains. But it could take up to five minutes for the doctors to type the same paragraph letter by letter. With speech recognition technology that understands African accents, doctors can take notes faster and save time. Oratunji said several hospitals in Nigeria, Kenya, Ghana and South Africa are now using Intron’s speech recognition software.
Nakunta Muwasu is a physician at Meridian Health Group in Nairobi, one of the hospitals using Intron Health’s software. “It makes taking short notes faster,” Mwas told TechCabal. But for long records, she prefers typing on the keyboard.
Intron has several papers under review and will present them at computer science conferences to showcase its research and products. In his one of a paper co-authored by Tobi Olatunji, Tejumade Afonja, and his five other computer scientists, the authors found that African accent models outperformed more common speech recognition software models. indicates that
Intron is one of the African startups using artificial intelligence to enable digital transformation. Earlier this year, Tunisian-born enterprise AI company InstaDeep, which was founded in 2015 and raised $107 million from investors, was acquired by BioNTech for $680 million. Intron, by contrast, has raised only $250,000 in rounds from family and friends to date since its founding in 2019. Nevertheless, the company was able to build and maintain the expensive infrastructure requirements of running an AI company, in part thanks to his partnership with GPU chip maker Nvidia.
As part of the same partnership, the company is currently running a $5,000 bounty hackathon inviting programmers to build better models using a small portion of the training dataset. Afri-Speech-200 is a developer challenge created in partnership with DSN, Masakhane, and Zindi to advance diversity in artificial intelligence. Lack of diversity and possible bias is one of the issues that AI practitioners and technology regulators must contend with. Intron’s ASR technology could prove to be a valuable tool for health workers in Africa, but lest AI systems be completely blind to situations outside the developed world where training data is plentiful. It also highlights the work you need to do to get there.
Also read:
