Anemia, a severe health condition affecting women of reproductive age and young children, continues to pose significant challenges in low-resource areas. The World Health Organization (WHO) recognizes this global health concern which has led to the establishment of the "Global nutrition target 2025-anemia," seeking to reduce anemia cases among women of reproductive age by 50%. To address this critical issue, researchers supported by the Department of Biomedical Engineering at the Indian Institute of Technology (IIT) Ropar and REAN Foundation, have developed an innovative solution - an AI-based chatbot for anemia screening. The time-efficient, cost-effective, and user-friendly screening tool is aimed at empowering communities with life-saving, early detection.

The research team has developed an AI-based chatbot for non-invasive and cost-effective anemia screening in remote areas using eye conjunctiva images captured from smartphones. It enables easy and accurate anemia screening without the need for extensive laboratory infrastructure or trained personnel. The chatbot, deployed on popular platforms like Telegram and WhatsApp, harnesses the power of artificial intelligence and machine learning to provide medical consultancy and disease diagnosis services. Utilizing a segmentation and classification model based on the REAN HealthGuru architecture, the chatbot provides reliable differentiation between anemic and non-anemic cases. The research has been published in the IETE Technical Review.

Dr. Ashish Sahani, Assistant Professor, Center for Biomedical Engineering (CBME) Indian Institute of Technology Ropar, said, “This innovative AI-based chatbot revolutionizes anemia screening, aiding mass screenings, telemedicine, and empowering frontline health workers. Its high accuracy enables effective self-diagnosis and early intervention. Future improvements include gender and age considerations, ensuring precision with a larger dataset. A significant breakthrough in telemedicine and public health screening."

For model development, The system uses advanced data preparation techniques, including data collection, quality checking, and structuring. A dataset comprising conjunctiva images along with gender, age, and hemoglobin (Hb) values was collected using Kobotoolbox and open-sourced eye images. The data was pre-processed, labeled, and augmented to enhance model performance. There are two essential models: a segmentation model to identify the Region of Interest (ROI) and a classification model to distinguish anemic from non-anemic cases. The segmentation model, based on a modified U-Net architecture with ResNet-34 as the encoder, achieved a mean Intersection Over Union (IOU) of 0.9272. The classification model, utilizing EfficientNet-B6 as the feature extractor, obtained an AUC of 0.970 for anemic classification.

To ensure the robustness of the chatbot, the researchers meticulously collected data from 160 individuals with anemia and 140 non-anemic individuals. This data served as the foundation for training the AI models, resulting in a powerful tool that can swiftly screen for anemia in just 35 seconds. The ease of use and accessibility of the chatbot make it a valuable resource for healthcare professionals, patients, and the general public alike.

"We are excited to introduce this cutting-edge AI-based chatbot, which has the potential to transform anemia screening in low-resource settings," stated Pallavi Sachdeva, senior AI engineer, REAN Foundation and the lead researcher of the project. "With its rapid results and user-friendly interface, the chatbot can be utilized by anyone, anytime, and anywhere, breaking barriers to healthcare access and facilitating early intervention."

“REAN Foundation envisions far-reaching impact by aligning this AI-based chatbot with the REAN HealthGuru platform, extending beyond individual screenings and presenting a significant breakthrough in telemedicine and public health screening. It can help governments in conducting mass screenings, leveraging the data to identify regions with high anemia prevalence, develop targeted awareness campaigns, and formulate long-term strategies for public health improvement. With its high accuracy, the chatbot will add to REAN HealthGuru platform’s capabilities as an effective tool for self-diagnosis and early intervention. Additionally, doctors can seamlessly integrate the chatbot into telemedicine consultations, empowering frontline health workers to collect valuable data for healthcare analysis.” said Sri Vasireddy, co-founder and CEO, REAN Foundation.

In future, the researchers envision further advancements in the model, exploring the inclusion of additional parameters such as gender, age, and ethnicity to enhance performance and accuracy; the bot could be trained on an extensive dataset in future that will allow it to effectively determine the severity of anemia and infer exact hemoglobin level. The team anticipates continued collaboration with like-minded researchers and engineers to further refine and expand the chatbot's capabilities.

The research team supported by Indian Institute of Technology Ropar and REAN Foundation invites fellow professionals to join the journey in transforming the landscape of anemia screening. By harnessing the potential of AI and digital infrastructure, we can work towards improving global health outcomes and save countless lives.

Read the published research in the IETE Technical Review.

Scientific summary of the project
This paper presents a comprehensive approach to develop an AI-based system for non-invasive and cost-effective anemia screening in remote areas using eye conjunctiva images captured from smartphones. The objective is to enable easy and accurate anemia screening without the need for extensive laboratory infrastructure or trained personnel. The dataset consists of 160 anemic and 140 non-anemic cases of varying ages and genders.v

For model development, The system uses advanced data preparation techniques, including data collection, quality checking, and structuring. A dataset comprising conjunctiva images along with gender, age, and hemoglobin (Hb) values was collected using Kobotoolbox and open-sourced eye images. The data was pre-processed, labeled, and augmented to enhance model performance. There are two essential models: a segmentation model to identify the Region of Interest (ROI) and a classification model to distinguish anemic from non-anemic cases. The segmentation model, based on a modified U-Net architecture with ResNet-34 as the encoder, achieved a mean Intersection Over Union (IOU) of 0.9272. The classification model, utilizing EfficientNet-B6 as the feature extractor, obtained an AUC of 0.970 for anemic classification.

It has the potential to aid mass screenings, facilitate telemedicine, and empower frontline health workers. The chatbot demonstrates high accuracy, making it an effective tool for self-diagnosis and early intervention. Future improvements may include considering gender and age for more precise classifications and a larger and diverse dataset to enhance performance further. Overall, the AI-based chatbot presents a significant breakthrough in telemedicine and public health screening