Michael A. Grasso

The integrality of speech in multimodal interfaces

A framework of complementary behavior has been proposed which maintains that direct-manipulation and speech interfaces have reciprocal strengths and weaknesses. This suggests that user interface performance and acceptance may increase by adopting a multimodal approach that combines speech and direct manipulation. This effort examined the hypothesis that the speed, accuracy, and acceptance of multimodal speech and direct-manipulation interfaces will increase when the modalities match the perceptual structure of the input attributes. A software prototype that supported a typical biomedical data collection task was developed to test this hypothesis. A group of 20 clinical and veterinary pathologists evaluated the prototype in an experimental setting using repeated measures. The results of this experiment supported the hypothesis that the perceptual structure of an input task is an important consideration when designing a multimodal computer interface. Task completion time, the number of speech errors, and user acceptance improved when interface best matched the perceptual structure of the input attributes.

Survey of handheld computing among medical students

The purpose of this study was to identify trends in the utilization and acceptance of handheld computers (personal digital assistants) among medical students during preclinical and clinical training. We surveyed 366 medical students and collected information on computer expertise, current handheld computer use, predicted future use, and user acceptance. Handheld computers were primarily used for personal applications by students during their preclinical training and as drug references and clinical calculators during their clinical training. In the future, all participants predicted they would use handheld computers at significantly higher rates and on a broader range of medical applications. The adoption of handheld computing was independent of user satisfaction. Those with more clinical experience were less satisfied with handheld computers, suggesting that the expectations of the more experienced users were not met. The lack of institutional support was seen as a key limitation.

Clinical applications of handheld computers

As the healthcare industry continues to become more distributed, healthcare organizations are increasing their reliance on mobile links to access patient information and to update their master database at the point of care. Handheld computers have evolved into a viable platform for these systems. While initial projects have shown promise, several questions remain. This article explores the unique characteristics of handheld computers with respect to user interface design and wireless access, and introduces a prototype development effort.

The long-term adoption of speech recognition in medical applications

This paper presents a survey on the long-term adoption of speech recognition in medical applications. Thirty-one participants who authored papers on medical speech recognition applications responded to the survey. The participants viewed speech technology more favorably today than when they originally published their papers. However, the adoption of speech applications did not always correspond with their enthusiasm. The survey suggested that hands-busy, eyes-busy, and mobility requirements are not always enough to offset current limitations in speech technology, There may need to be other benefits, such as decreased medical costs and increased quality of care, or other factors, such as using a limited vocabulary.

Feasibility study of voice-driven data collection in animal drug toxicology studies

The object of this study was to determine the feasibility of using voice recognition technology to enable hands-free and eyes-free collection of data related to animal drug toxicology studies. Specifically, we developed and tested a prototype voice-driven data collection system for histopathology data using only voice input and computer-generated voice responses. The overall accuracy rate was 97%. Additional work is needed to minimize training requirements and improve audible feedback. We conclude that this architecture could be considered a viable alternative for data collection in animal drug toxicology studies with reasonable recognition accuracy.

Task integration in multimodal speech recognition environments

A model of complementary behavior has been proposed based on arguments that direct manipulation and speech recognition interfaces have reciprocal strengths and weaknesses. This suggests that user interface performance and acceptance may increase by adopting a multimodal approach that combines speech and direct manipulation. More theoretical work is needed in order to understand how to leverage this advantage. In this paper, a framework is presented to empirically evaluate the types of tasks that might benefit from such a multimodal interface.

Structured speech input for clinical data collection

This paper presents an environment for collecting clinical data using structured speech input. The system uses nomenclature terms and data consistency rules to limit the speech input to a discrete set of phrases which can be configured to support a wide range of studies. This information is stored in an event-oriented data model as strongly typed observations for subsequent distribution and data analysis. The use of speech input should result in increased efficiency during hands-busy data collection tasks. The strongly typed data, as opposed to free-form narratives, ensures a quantitative information base for analysis and clinical decision support.

Detection of unsafe action from laparoscopic cholecystectomy video

Wellness and healthcare are central to the lives of all people, young or old, healthy or ill, rich or poor. New computing and behavioral research can lead to transformative changes in the cost-effective delivery of quality and personalized healthcare. Also beyond the daily practice of healthcare and wellbeing, basic information technology research can provide the foundations for new directions in the clinical sciences via tools and analyses that identify subtle but important causal signals in the fusing of clinical, behavioral, environmental and genetic data. In this paper we describe a system that analyzes images from the laparoscopic videos. It indicates the possibility of an injury to the cystic artery by automatically detecting the proximity of the surgical instruments with respect to the cystic artery. The system uses machine learning algorithm to classify images and warn surgeons against probable unsafe actions.

Prescriptions Written for Opioid Pain Medication in the Veterans Health Administration Between 2000 and 2016

Objectives: The purpose of this study was to identify national opioid pain medication (OPM) prescribing trends within the Veterans Health Administration (VA), and assess the impact of educational campaigns introduced in 2010 and 2013. Methods: We created a national cohort that documents more than 21 million patient records and 97 million outpatient OPM prescriptions covering a 17-year period. We examined OPM prescriptions in emergency departments, outpatient clinics, and inpatient settings. Results: The cohort accounted for 2.5 billion outpatient clinic visits, 18.9 million emergency department visits, and 12.4 million hospital admissions. The number of OPM prescriptions peaked in 2011, when they were provided during 5% of all outpatient visits and 15% of all emergency department visits. The morphine milligram equivalents (MMEs) peaked in 2014 at almost 17 billion in outpatient clinics and at 137 million in emergency departments. In 2016, OPM prescriptions were down 37% in outpatient clinics and 23% in emergency departments, and MMEs were down 30% in both settings. Prescriptions for hydrocodone and tramadol increased markedly between 2011 and 2015. OPM doses in inpatient settings continued to rise until 2015. Conclusions: We used a large national cohort to study trends in OPM prescriptions within the VA. Educational efforts to reduce the number of OPM prescriptions coincided with these reductions, but were initially associated with an increase in OPM dosage, an increase in the use of tramadol and hydrocodone, and an increase in the use of OPMs in inpatient settings.

Integrating domain knowledge in supervised machine learning to assess the risk of breast cancer

We used various supervised machine learning and data mining techniques to generate a model for predicting risk of breast cancer in post menopausal women using genomic data, family history, and age. In this paper, we propose an approach to select nine best SNPs using various feature selection algorithms and evaluate binary classifiers performance. We have also designed an algorithm to incorporate domain knowledge into our machine learning model. Our observations revealed that the machine learning model generated using both the domain knowledge and the feature selection technique performed better compared to the naive approach of classification. It is also interesting to note that, in addition to selecting nine best SNPs, feature selection resulted in removing age from the set of features to be used for cancer risk assessment.