March 11-12, 2023, Virtual Conference
Tae Hyun Kim, Won Seok Jan, Sun Cheol Heo, Min Dong Sung and Yu Rang Park, Department of Biomedical Systems Informatics, Yonsei University College of Medicine, Seoul, South Korea
Federated learning (FL) has been used for model building across distributed clients. However, FL cannot leverage vertically partitioned features to increase the model complexity. In this study, we proposed a personalized progressive federated learning (PPFL) model, which is a multi-model PFL approach that allows the leveraging of vertically partitioned client-specific features. The performance of PPFL was evaluated using two datasets: the Physionet Challenges 2012 dataset and a real-world dataset composed of eICU data and the Severance Hospital, Seoul, South Korea. We compared the performance of inhospital mortality and length of stay prediction between our model and the FedAvg, FedProx, and local models. The PPFL showed an accuracy of 0.849 and AUROC of 0.790 in in hospital mor-tality prediction, which are the highest scores compared to client-specific algorithm. For length-of-stay prediction, PPFL also showed an AUROC of 0.808 in average which was the highest among all comparators.
Personalized Federated Learning, Vertical Federated Learning, Non-IID data.
Homa Hosseinmardi1, AmirGhasemian2, ShrikanthN arayanan3, KristinaLerman3, EmillioF errara3, 1University of Pensylvania, Department of Computer and Information Sciences, Philadelphia, PA, USA, 2Yale University, Social Science Department, New Haven, CT, USA and 3 Information Sciences Institute, University of Southern California, CA, USA
Todays densely instrumented world offers tremendous potential for continuous acquisition and analysis of multimodal sensor data, allowing for temporal characterizing individual behavior. Can rich sensory data be efficiently coupled with predictive modeling techniques to provide contextual and insightful assessments of individual behavior? The data are noisy, incomplete, and derived from multiple sources, each with a different temporal resolution and dimension. Moreover, longitudinal studies typically target multiple aspects of human behavior, such as well-being, performance, and personality, which complicates handcrafted feature engineering. In this paper we propose an algorithm for joint decomposition of input and target variables in multimodal data, supervised tensor embedding (STE), in order to meet these challenges. The latent features obtained from STE may be fed into any regression model for the estimation of the target variable(s). It is possible to combine features originating from different sources after embedding high-order data. We further demonstrate that feature selection on higher-order data can enhance performance. Two real-world datasets with 29 different target variables are used to evaluate the efficiency of our method. In comparison to the state of the art, our method outperforms other methods on 21 prediction tasks. There are, however, some aspects of human behavior that are unpredictable regardless of the method, which may be due to the limitations of multilinear methods or the absence of appropriate independent variables.
Supervised tensor embedding, Multivariate time series regression, Human behavior, Wearable devices.
Christopher Komen1*, Dr Patrick Onyango2, Dr.Samuel Bonuke3, 1PG Student: School of Public Health and Community Development, Maseno University, Kenya, 2Lecturer, School of Physical and Biological Sciences, Department of Zoology, Maseno University, Kenya, 3Lecturer, School of Medicine, Department of Medical Biochemistry, Maseno University, Kenya
Purpose: This study seeks to investigate factors influencing the utilization of the International Quality Care (IQCare) system for Clinical Decision Support (CDS) in Kenya. The study was carried out with 81 health workers from 13 sampled health facilities where IQCare system had been deployed by January 2014. Structured questionnaires and focus group discussion guides were used for data collection. Findings: The study found that the differences in age, gender, staff cadre and education level were not associated with differences in utilization of IQCare system for CDS. However, significant association between human resource availability and utilization of IQCare for was found; with IQCare training, availability of mentorship significantly associated with use of IQCare for CDS. Linking computers via local area network, challenges of internet, and availability of alternative power back-up were less significant barriers. The study recommends improving staffing, training and mentorship of staff in IQCare systems to improve its utilization.
Clinical decision support (CDS), Electronic medical record (EMR), Infrastructure, International Quality Care (IQCare), User perspective and human resource availability.
Biniam Gedlie Lakew1 and Solomon Negash2, 1School of Information Science, Addis Ababa University, Ethiopia, 2Department of Information Systems and Security, Kennesaw State University, USA
The implementation of a health information management system, which supports and provides for medical facilities and patients, enables basic public health services. Despite this, national and local health managers can track and evaluate health services and evaluate the success of health initiatives using HMIS data. Additionally, when managers actively participate in the adoption of HMIS, they will be able to properly describe, monitor, and assess system performance among themselves. The study's primary objective is to draw lessons from the HMIS implementation at a Public Hospital in Ethiopia. This study is also intended to identify the implementation of the HMIS and its factors that specifically on three specialized teaching hospitals in Addis Ababa. The researcher will conduct in this research questionnaires and interviews as a primary source that will prepare and distribute to a sample of a population under consideration—in addition, analyzing the data through utilizing the quantitative and qualitative methods requiring to standardize information and about the subject being examine by collecting data from secondary sources. The study will identify significant factors and provide implementation guidance for health systems that may be used by all public hospitals when implementing health management information systems. The findings of this study are not only relevant to hospitals in Ethiopia, but they may also be applied in other countries.
eHealth, eHealthcare, Health Information Systems (HIS), Health Management Information Systems (HMIS)
Hafsa Gulzar1, Jiyun Li1, Arslan Manzoor2, Sadaf Rehmat3, Usman Amjad4 and Hadiqa Jalil Khan4, 1Computer Science and Technology, Donghua University, Shanghai, China, 2Mathematics and Computer Science, University of Catania, Catania, Italy, 3Computer Science Department, PIEAS, Islamabad, Pakistan, 4Computer Science and Information Technology, Islamia University of Bahawalpur Pakistan
With the development of computer -systems that can collect and analyze enormous volumes of data, the medical profession is establishing several non-invasive tools. This work attempts to develop a non-invasive technique for identifying respiratory sounds acquired by a stethoscope and voice recording software via machine learning techniques. This study suggests a trained and proven CNN-based approach for categorizing respiratory sounds. A visual representation of each audio sample is constructed, allowing resource identification for classification using methods like those used to effectively describe visuals. We used a technique called Mel Frequency Cepstral Coefficients (MFCCs). Here, features are retrieved and categorized via VGG16 (transfer learning) and prediction is accomplished using 5-fold cross-validation. Employing various data splitting techniques, Respiratory Sound Database obtained cutting- edge results, including accuracy of 95%, precision of 88%, recall score of 86%, and F1 score of 81%. The ICBHI dataset is used to train and test the model.
Machine learning, convolutional neural networks (cnn), transfer Learning, cross validation, mfcc, vgg16.