mHealth Apps

Dengue has been endemic in the country since 2000. The national guidelines provide methods for IV fluid administration based on patient parameters, but the calculation being complex and time-consuming, poses challenges to healthcare professionals during peak dengue seasons when hospitals become overburdened with dengue patients. To address this issue, we propose DengueDrops, a web-based calculator for the calculation of the required fluid amounts and IV flow rates for non-shock (group-B) dengue patients.

The recent 'Mpox' outbreak, formerly known as 'Monkeypox', has been a significant public health concern and spread to over 110 countries globally. Mpox can be difficult to recognize in its early stages because its skin lesions often resemble those caused by other infectious diseases. In our lab, we develop web-based AI systems that help make this kind of screening faster, more accessible, and more scalable. Our work combines medical image analysis, dataset development, and deep learning to identify visual patterns in skin lesion photographs that may indicate suspected mpox. We also study how to improve fairness and generalizability across diverse skin tones, since skin appearance can vary substantially across populations and biased systems can reduce real-world usefulness.

Beyond algorithm development, we translate our models into practical prototype web applications where users can upload skin images for rapid preliminary assessment. These tools are not intended to replace laboratory testing or expert clinical judgment. Instead, they are designed to serve as decision-support systems that can assist with early screening, outbreak preparedness, and triage in settings where specialist care or confirmatory testing may not be immediately available.

Publications

• https://www.sciencedirect.com/science/article/pii/S1746809424008000
• https://arxiv.org/abs/2207.03342

Diagnosing diseases in resource-constrained healthcare settings is often challenging because clinicians must consider many possible conditions from limited patient information while working under significant time and resource constraints. In tropical regions, illnesses such as dengue, malaria, and chikungunya frequently present with overlapping symptoms, making accurate differential diagnosis difficult.

In our lab, we explore AI-driven approaches to support clinical decision-making in such environments by developing lightweight medical language models tailored for healthcare applications. Our research focuses on adapting large language models with domain-specific medical knowledge while reducing computational requirements so they can operate in resource-limited settings. By integrating symptom descriptions, medical knowledge, and probabilistic reasoning, these systems aim to assist clinicians in prioritizing likely diagnoses during patient evaluation.

Dengue has become a major public health challenge in many low-resource settings, yet effective vector surveillance still depends heavily on expert entomologists and laboratory equipment. In our lab, we develop AI-based tools that make mosquito identification faster, more scalable, and more accessible using ordinary smartphone cameras. By combining field-collected insect image datasets, expert annotation, and deep learning models, we create systems that can distinguish dengue-carrying mosquito species from visually similar insects with high accuracy.

Our work aims to reduce the need for specialized expertise in routine surveillance and enable real-time, field-ready screening through user-friendly digital platforms. Beyond species identification, we are also interested in integrating these tools with spatial mapping and public health workflows so that mosquito activity can be monitored more effectively and control measures can be deployed where they are needed most.