PhD Defense: "Multimodal Analytics for Healthcare"

Carlos Torres

November 20th (Monday), 1:00pm
NOTE: location moved to Harold Frank Hall, Room 4614 (ECE Conf. Rm.)

The ailing healthcare system demands effective, autonomous solutions to improve services and provide individualized care, while reducing the burden on the scarce healthcare workforce. Most of these solutions require a multidisciplinary approach that combines healthcare with computational abilities. Intensive Care Unit (ICU) rooms are of particular interest due to their strategic importance, workflow controls, and potential for dissemination of clinical findings and developments. The work presented in this thesis introduces a multimodal, multiview sensor network along with methods and solutions to monitor mock-up and real medical ICU rooms. One prominent outcome of this work includes enabling the medical analysis of preventable ICU conditions such as sleep disorders, decubitus ulcerations, and hospital-acquired infections. Some of the challenges include illumination variations and partial and complete occlusions, such as blankets or privacy curtains. In addition, proper monitoring of human environments requires person identification, which is prohibited by healthcare privacy-protection stipulations and can be limited by scene constraints. The problems tackled include patient-pose classification, pose-motion analysis and summarization, role representation and identification, and activity and event logging. These problems are addressed via a non-intrusive, non-disruptive, multimodal, multiview sensor network (i.e., Medical Internet-of-Things). The multimodal data is combined with coupled optimization to estimate source weights and accurately classify patient poses. Pose transitions are represented using deep convolutional features and pose durations are modelled via segments. The described techniques serve to differentiate between poses and pseudo-poses (transitions) and create effective motion summaries. Role representations are tackled using novel appearance and semantic interaction maps to assign generic labels to individuals (e.g. doctors, nurses, or visitors) without using identifiable information (e.g., face tracking or badges), which is prohibited in healthcare applications. Finally, activities and events are analyzed using contextual aspects, where aspect bases and weights are learned and then used to classify activities. The objective of this thesis is to enable the development, evaluation, and optimization of individualized therapies, standards-of-care, infrastructural designs, and clinical workflows and procedures.

About Carlos Torres:

photo of carlos torresCarlos graduated from San Jose State University (SJSU) with a dual Bachelor of Science degree in Electrical Engineering and BioEngineering. During his undergraduate career, Carlos worked as a Research Assistant at Hewlett-Packard Labs in Palo Alto, where he helped develop Paper-Thin Flexible-Semi-Conductor Materials under the supervision of Drs. Karl Taussig and Warren Jackson. He also worked as a Research Assistant in the Egger's Lab at SJSU, where he investigated BioMaterials. His desire to pursue a graduate degree developed while participating in the SURE program, established by Dr. Gary May, at the Georgia Institute of Technology (GaTech). At GaTech, Carlos joined the Healthcare Robotics Lab as a research assistant to Dr. Charles C. Kemp and was mentored by Cressel D. Anderson and Travis Dayle (Google X and Cobalt Robotics). His undergraduate work was published in BioChemistry Journals and Presented at a BioEngineering Conference (NIH-ABRCMS).

Carlos obtained a Master’s of Science degree from the Department of Electrical and Computer Engineering at the University of California Santa Barbara, where he joined the Vision Research Laboratory (VRL) under the supervision of Professor B. S. Manjunath. In the VRL, Carlos gained essential experience in planning and conducting independent research that ranged from grant proposal drafting to publishing and presenting scientific findings, and participating in conferences, workshops, journal reviews, and events. Carlos’ work focuses on the development of multimodal analysis methods and algorithms for healthcare.

Carlos is an active member in the tech community in Santa Barbara County. He worked as a Software Developer for Caugnate in Goleta, as a Lead Data Scientist for Social Intelligence (now Carpe Data) in Santa Barbara, and as a Machine Learning and Data Sciences Researcher for Procore Technologies in Carpinteria where his efforts are focused on helping to automate and optimize construction management.

Hosted by: Professor B. S. Manjunath