Awards

Tavpritesh Sethi of Indraprastha Institute of Information Technology Delhi in India will develop an AI-based platform for AMR surveillance and management across a broad network of public and private hospitals in India. The platform will extract weekly data on AMR from the All India Institute of Medical Sciences, New Delhi (AIIMS Delhi) hospital and the Max Healthcare hospital network, including patterns of antibiotic prescriptions across the network. It will use a federated data analysis approach (joint analysis without sharing the data itself), and they will develop and integrate AI-based models to identify and predict trends in AMR. They will also create applications driven by these models to widely and effectively communicate the analyses to healthcare professionals. This will support antibiotic stewardship and data-driven AMR management at both the local and regional levels.

Sai Raj Reddy of Daia Tech Private Limited in India will develop a program to increase access to health education and related resources for adolescent girls in rural areas of the Indian state of Karnataka. The program will be developed in partnership with the Karnataka Health Promotion Trust, building on their ongoing work with local schools, healthcare providers, community leaders, and government agencies. After engaging community members to understand the local context, they will develop resources for adolescent girls including life skills courses and health education workshops and pilot test the program in selected villages. They will integrate AI tools across the program to broaden participation and to broaden the range of health outcomes improved for adolescent girls.

Belehalli Pavan of Strideaide Private Limited in India will develop an AI-based platform to increase timely access to diagnosis and treatment for diabetic foot ailments. Early detection of peripheral neuropathy, peripheral arterial disease, and diabetic foot ulcers would enable interventions that reduce the need for limb amputation. The platform will build on their existing network of podiatry clinics located in a variety of public spaces. These clinics are staffed with a paramedic providing treatment guided by diagnostic tools including a foot mat with a plantar pressure sensor to help predict and assess foot-bottom ulceration. Existing and new digital data from these clinics will be used to build a diabetic foot registry, to improve automated assessment through AI-based analysis, and to train an LLM with a chatbot interface for clinical decision support across the podiatry clinics.

Amrita Mahale of ARMMAN in India will incorporate an LLM-based chatbot into the Kilkari mobile health service to answer questions about pregnancy and infant care. Kilkari currently provides weekly pre-recorded messages on preventive care, and it is implemented in partnership with India's Ministry of Health and Family Welfare. They will assess different available LLMs and train the best candidate with the Kilkari database of vetted content. They will pilot test the model with Kilkari users through WhatsApp, engaging users in Delhi and in the state of Jharkhand to encompass diverse participants spanning urban, rural, and tribal populations. The test will include regular surveys to measure the service's accuracy, relevance, and usability.

Mohanasankar Sivaprakasam of the Healthcare Technology Innovation Centre at the Indian Institute of Technology Madras in India will develop an AI-based platform to support diagnosis and report generation for endoscopic gastrointestinal exams. They will use curated datasets of annotated endoscopic images to develop an AI-based model for diagnosing gastrointestinal abnormalities, such as polyps and ulcers. The data will also be used to train a Large Language Model (LLM) to generate diagnostic reports of endoscopic exams including representative images and text descriptions. This will improve report quality and consistency for better diagnosis and more accurate, detailed patient records, with the report automation reducing the time and personnel required. Together, these platform components will improve patient care in gastroenterology, including more efficient care across more patients.

Ntombifikile Mtshali of Shout-It-Now with Elona Toska of the University of Cape Town, both in South Africa, will pilot test use of ChatGPT to provide information on sexual and reproductive health that helps young women and adolescent girls in South Africa make informed decisions and effectively access health services. They will test integration of the chatbot into Shout-It-Now's existing platforms: a tablet-based platform in mobile clinics staffed by young women and a mobile phone app. They will train ChatGPT to provide information on sensitive topics, including gender-based violence, HIV infection risk, and pregnancy, using existing materials and guided by a workshop with mobile clinic staff. Users' perception of the chatbot and the chatbot's effectiveness in increasing the use of health services will be assessed using on-line questionnaires and phone surveys across five demographically different districts.

Vijay Sai Trap of OnionDev Technologies Pvt. Ltd. in India will develop an AI-based platform to provide accurate and private automated answers to questions on sexual health and mental well-being for youth and women in rural India. They will generate a dataset of community-generated questions on these topics during a mental health awareness campaign in the Indian states of Uttar Pradesh, Madhya Pradesh, and Tamil Nadu. With the support of subject matter experts, they will add answers to these questions from the relevant literature. They will then create a master dataset of questions and answers, including translations using an existing LLM trained on local Indian languages. This dataset will be used to compare different AI-based models to identify the one best able to effectively answer questions on these sensitive topics.

Himanshu Sinha of the Indian Institute of Technology Madras in India will develop an LLM-based chatbot to provide personalized reliable guidance on antenatal care in multiple Indian languages, particularly for mothers without regular access to health care. The chatbot will use an open-source LLM and will be incorporated into a mobile phone application. The project will begin by surveying pregnant women seeking care in a variety of health care settings and across all three trimesters, asking them what features they would want in a pregnancy app. The LLM will be trained with information from textbooks, clinical manuals, government health resources, and guidelines from professional organizations. Initially the chatbot will function in Hindi and Tamil, with additional Indian languages to follow. The app will track pregnancy milestones and deliver relevant evidence-based advice.

Anurag Agrawal of Ashoka University in India will develop an LLM-based platform to support medical decision making by home healthcare workers in India who are meeting the growing demand for home-based palliative care. The platform will use an existing proprietary LLM to extract and summarize relevant clinical information, connecting it with an existing open-source AI chatbot to generate advice in a conversational format for healthcare workers. They will test the platform using a dataset they will build of palliative care scenarios, focused initially on care for lung diseases, and they will compare outputs from several different open-source LLMs to guide the platform's final configuration. Expert clinicians will evaluate the clinical advice generated by the platform for its factual accuracy and relevance to the Indian sociocultural context.

Sona Shah of Neopenda, PBC in Kenya will integrate ChatGPT-4 as a virtual assistant for the wearable, vital sign monitor neoGuard, supporting healthcare providers in effectively monitoring and managing neonatal health. They will train ChatGPT-4 to help providers identify and address challenges with the neoGuard monitor, such as poor sensor placement on the patient, and to give providers appropriate recommendations based on vital sign data together with the clinical information they gathered. This real-time clinical decision support would be particularly beneficial in remote and understaffed healthcare facilities. For model training, they will use a dataset of newborns admitted to a hospital in Kenya, including vital signs, clinical histories, and treatment outcomes, as well as insights from unstructured clinical notes extracted using natural language processing. They will evaluate use of neoGuard with ChatGPT-4 for reliability, accuracy, and user-friendliness, and compare neonatal patient outcomes before and after ChatGPT-4 integration with the monitor.