48440 research outputs found
Sort by
Neural Network Models for Assessing Cardiac Health
No full textHypertrophic cardiomyopathy (HCM) is a common inherited cardiomyopathy with heterogeneous morphology and clinical trajectories characterized by elevated risks of atrial fibrillation (AF), heart failure (HF), and sudden cardiac death (SCD). Echocardiography is the most widely available imaging modality for HCM assessment. However, clinical interpretation of echocardiography video is time-consuming and subject to inter-reader variability. This dissertation develops and evaluates a cohesive, three-thrust Artificial Intelligence (AI) framework that progresses from echocardiography video-based detection to multimodal phenotyping and longitudinal time-to-event risk stratification. Across research thrusts, the learned video representations serve as the connective tissue: embeddings learned for detection are reused downstream for phenotyping and prognostic modeling. In Thrust 1, state-of-the-art Video Action Recognition (VAR) neural network architectures are adapted to cardiac function assessment and HCM detection from echocardiogram videos. First, VAR models for ejection fraction (EF) classification are investigated on the EchoNet-Dynamic dataset in order to establish feasibility and compare multiple spatiotemporal backbones. Next, an end-to-end HCM detection framework leveraging two-pathway temporal modeling and transfer learning is introduced, which is then further improved for robustness using deep ensemble fusion. In Thrust 2, building on the echocardiography-derived embeddings learned in Thrust 1, HCM-PhenotypeNet, a multimodal phenotyping framework that integrates echocardiography-derived embeddings with more than 100 structured clinical variables to discover clinically meaningful HCM phenogroups, is proposed. Using a cohort of 156 HCM patients and 1,553 echocardiography video clips, dimensionality reduction and deep embedded clustering is performed along with validation of cluster quality using internal indices and characterization of phenogroups using statistical comparisons and association rule mining for interpretability. In Thrust 3, using these echo-derived embeddings as additional covariates alongside structured clinical variables, a unified, reproducible survival modeling pipeline is developed to operationalize time-to-event endpoints for incident AF, advanced HF, and SCD-related events and benchmark classical and neural survival models under a consistent evaluation protocol. Across endpoints, the results highlight strong performance of tree-based survival ensembles as competitive baselines and quantify the incremental value of adding echocardiography embeddings to clinical covariates for longitudinal risk stratification. Collectively, this work demonstrates that spatiotemporal learning from echocardiography video can support scalable, automated HCM detection, that multimodal embeddings can reveal clinically coherent phenogroups, and how these representations can be incorporated into time-to-event modeling to support individualized risk estimation. Collectively, these three research thrusts form a unified framework that progresses from video-based cardiac function assessment, to multimodal disease phenotyping, and finally to longitudinal risk stratification and time to adverse event prediction, enabling a comprehensive AI framework for HCM assessment
Suppression of Thermal Decomposition in Woody Biomass by Hig
No full textThis research introduces an experimental approach to suppressing thermal decomposition in woody biomass. Unlike conventional fire-suppression studies, this method utilizes high-viscosity fluid coatings to delay pyrolysis during radiant heat exposure. Xanthan gum and methylcellulose solutions in water were used in an ongoing study. This study focuses on sodium polyacrylate solutions. These materials are distinct for their rheological, moisture-retention, and thermal response properties. The project was conducted in collaboration with the Combustion Engineering Laboratory at the Shibaura Institute of Technology, where research focuses on flame propagation and pyrolysis behavior
Enhancing API Usability Via MCP Context Delivery for LLMs
No full textThe role of MCP servers & LLMs in large APIs were explored with the sponsorship of Quantifi Solutions. An MCP server was created that queried a large Quantifi API dataset and generated code using the Quantifi libraries. This MCP server was also used to explain how bond pricing models can be implemented using the Quantifi API. A chatbot was developed that prompted the user for a query, used the MCP server to obtain context about the API and generated Python code that used Quantifi's proprietary classes & bond pricing methods. The chatbot was successfully created to generate a working script that can price bonds with a SOFR curve. The model produced a working code template that follows Quantifi’s intended usage patterns & could be adjusted to fit specific pricing or modeling requirements
Automatic Storage and Retrieval System
No full textThis project develops an autonomous mobile robot for warehouse automation, replacing a rigid rail-based system with flexible mecanum-wheel mobility. The robot navigates grid-based warehouse layouts, plans paths around obstacles, and retrieves items from multi-level shelving without human intervention. Developed through WPI and Kyoto University of Advanced Science's international collaboration, the system successfully demonstrates autonomous navigation and intelligent shelf selection with occupancy detection, and serves as an extensible foundation for future warehouse robotics development
A Plan for Water Refill Stations at University of the Aegean
No full textAccess to safe drinking water on Syros, a Greek island in the Cyclades, is limited. Working with the University of the Aegean, this project assessed students’ willingness to shift from bottled to filtered tap water on campus. Through a multi-method approach, we found students were unlikely to shift due to their distrust, lack of water quality data, and limited accessibility to refill stations. We then developed recommendations to expand and promote refill stations to reduce plastic waste and increase sustainability
Building BioTech Cities: Impact of Living Support Services
No full textIn this project, a study is conducted on a business incubator subsidiary, Transfar Science & Technology City. The project focuses on examining living support systems such as housing, supermarkets, networking events and more. The team applied various data collection methods, such as literature studies, interviews, and surveys to formulate correlations between the entrepreneurial market around science parks, and highly sought-after conveniences by job seekers, in the U.S. and China. The team offers informed conclusions, advising which supports to develop further to continue to attract overseas returnees. Additionally, they found providing increased residential spaces, supermarkets, and nearby casual meeting spaces can help attract young talent and overseas returnees to an innovation park
Preserving Japan’s Indigo Tradition
No full textThis study examines current challenges and evolving approaches to the continuation of indigo dyeing in Japan. Through semi-structured interviews, our findings revealed growing interest in aizome at both professional and amateur levels. We also explored innovative efforts by artisans to adapt traditional practices through workshops and new commercial applications. The study further explores the feasibility of community-based collaboration as a strategy for long-term continuity, highlighting both its potential and its limitations. This research demonstrates that the future of aizome depends not only on individual artisans, but on interconnected networks of farmers, artisans, and communities working collectively to carry the tradition
Touristification in Kyoto
No full textThe number of tourists visiting Kyoto city has been steadily increasing, potentially having a detrimental effect on the city and its residents. Previous studies have examined overtourism in Kyoto, but few have considered lodging, shops, and restaurants together. In this project, we compiled indicators from existing research, adapted them for Kyoto, and applied them to selected neighborhoods to evaluate tourist accommodation facilities, shops, and restaurants. Our indicators distinguished highly tourist-serving streets from areas where they remained more local-serving. Our adapted indicators provide a useful framework for analyzing tourism’s impacts in Kyoto. Future studies can use our indicators to continue to develop a model that analyzes Kyoto in its entirety
UNOPS: Assessing Visitor Satisfaction in Albania
No full textFollowing the conclusion of the United Nations Office for Project Services (UNOPS) Albania heritage revitalization project and the handover of the National Puppet Theatre (NPT) and the Ethnographic Museum of Kavajë (EMK) to their respective caretakers, UNOPS sought final feedback on the renovation process and results, along with recommendations to guide future projects. Thus, a thorough study of the sites was conducted focusing on ascertaining levels of satisfaction among visitors and assessing the accessibility of both sites. A mixed methodology was used, including semi-structured interviews, surveys, and observations. Using the collected data, the research team generated a measure of impact of UNOPS’ renovations and provided the caretakers of NPT and EMK with actionable recommendations
Developing Learning Resources for Mālama Learning Center
No full textThis project develops StoryMaps for Mālama Learning Center to promote understanding of their four restoration sites in Leeward O’ahu. The sites are Pālehua, Nānākuli Muliwai, Kaʻala Kīpuka, and Awāwalei Food Forest. Our team worked with MLC staff at each location, which allowed us to represent the history, ecological conditions, and community involvement for each site. This information was shown in StoryMaps that support education and outreach by making each site’s purpose, challenges, and progress easy to see. Together, they highlight the cultural and ecological importance of the land while documenting the efforts led by MLC and community partners. The completed maps provide resources that strengthen engagement and learning the story of restoration for areas in Leeward O’ahu