Gupta collaborative research to improve the quality of mHealth datasets for AI/ML applications in HCT research

health care worker and smart watch, mHealth data

Allogeneic hematopoietic cell transplantation (HCT), often called a stem cell transplant, is a powerful treatment. It replaces the malfunction or cancerous bone marrow for patients suffering from leukemia, lymphoma, sickle cell disease, and other high-risk blood diseases.

The treatment requires a lengthy recovery, facilitated by a committed caregiver. Their work is intense, and often isolating. They can face psychological, social, and physical risks that not only affect the caregiver, but ultimately the health of the patient.  With the emergence of mobile health technologies and widespread, routine use of wearables, this person-generated health data has become a promising source for biomedical research.

Artificial intelligence and machine learning (AI/ML) techniques, however, are difficult to apply in HCT research. Data quality issues like high dimensional data, diverse data types, dynamic evolution of disease states, lack of labelled data, frequent and irregular data sparsity, and data integration issues complicate applying AI/ML.

Vibhuti Gupta

Vibhuti Gupta, Ph.D., assistant professor of computer science and data science, is representing Meharry SACS on a collaboration with the University of Michigan to develop high quality mHealth dataset for AI/ML applications in HCT research, and to share that processed data and developed methods with the research community. An NIH (NHLBI) division supplemental grant is funding their work.

“Mobile health, or mHealth, data holds promise in early diagnosis of post-HCT complications ,” says Dr. Gupta, who explores mHealth data in the mHealth Wearable Sensors Lab.

Affordable, wearable sensors can be combined with novel data analytics to both collect and interpret multi-parameter mHealth data, such as sleep, heart rate, steps and other physical activity.

“They create rich streams of data that, with the application of AI/ML analytics, have the potential to lead to predicting health risks early. However, these datasets are highly unstructured, complex, and messy. That is because they are generated continuously and at a high frequency, with thousands of observations per second,” says Dr. Gupta.

Thus, there is an urgent need to develop novel preprocessing procedures, enhancing data quality and data readiness for AI/ML applications in HCT research.

During a previous clinical trial, the University of Michigan collected mHealth data, clinical data, and other data from patient reported outcomes and surveys. Those raw datasets are stored in secure, HIPAA-compliant server databases. Now, they need to be extracted, preprocessed, and integrated in preparation for applying AI/ML techniques.

“The NIH has funded this collaboration between the University of Michigan and Meharry Medical College to improve the AI/ML readiness from the data collected in that trial,” says Dr. Gupta.

The collaboration includes expertise in biomedicine, data management, and AI/ML to make NIH-supported data useful and usable for AI/ML analytics. Their work will focus on two aims, 1) developing high quality mHealth datasets for AI/ML applications in HCT research; and 2) sharing the processed data and developed methods for AI/ML and the mHealth research community.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *