Grand Challenges at the Interface of Engineering and Medicine

Over the past two decades Biomedical Engineering has emerged as a major discipline that bridges societal needs of human health care with the development of novel technologies. Every medical institution is now equipped at varying degrees of sophistication with the ability to monitor human health in both non-invasive and invasive modes. The multiple scales at which human physiology can be interrogated provide a profound perspective on health and disease. We are at the nexus of creating "avatars" (herein defined as an extension of "digital twins") of human patho/physiology to serve as paradigms for interrogation and potential intervention. Motivated by the emergence of these new capabilities, the IEEE Engineering in Medicine and Biology Society, the Departments of Biomedical Engineering at Johns Hopkins University and Bioengineering at University of California at San Diego sponsored an interdisciplinary workshop to define the grand challenges that face biomedical engineering and the mechanisms to address these challenges. The Workshop identified five grand challenges with cross-cutting themes and provided a roadmap for new technologies, identified new training needs, and defined the types of interdisciplinary teams needed for addressing these challenges. The themes presented in this paper include: 1) accumedicine through creation of avatars of cells, tissues, organs and whole human; 2) development of smart and responsive devices for human function augmentation; 3) exocortical technologies to understand brain function and treat neuropathologies; 4) the development of approaches to harness the human immune system for health and wellness; and 5) new strategies to engineer genomes and cells.

aEYE: A deep learning system for video nystagmus detection

Background

Nystagmus identification and interpretation is challenging for non-experts who lack specific training in neuro-ophthalmology or neuro-otology. This challenge is magnified when the task is performed via telemedicine. Deep learning models have not been heavily studied in video-based eye movement detection.

Methods

We developed, trained, and validated a deep-learning system (aEYE) to classify video recordings as normal or bearing at least two consecutive beats of nystagmus. The videos were retrospectively collected from a subset of the monocular (right eye) video-oculography (VOG) recording used in the Acute Video-oculography for Vertigo in Emergency Rooms for Rapid Triage (AVERT) clinical trial (#NCT02483429). Our model was derived from a preliminary dataset representing about 10% of the total AVERT videos (n = 435). The videos were trimmed into 10-sec clips sampled at 60 Hz with a resolution of 240 × 320 pixels. We then created 8 variations of the videos by altering the sampling rates (i.e., 30 Hz and 15 Hz) and image resolution (i.e., 60 × 80 pixels and 15 × 20 pixels). The dataset was labeled as "nystagmus" or "no nystagmus" by one expert provider. We then used a filtered image-based motion classification approach to develop aEYE. The model's performance at detecting nystagmus was calculated by using the area under the receiver-operating characteristic curve (AUROC), sensitivity, specificity, and accuracy.

Results

An ensemble between the ResNet-soft voting and the VGG-hard voting models had the best performing metrics. The AUROC, sensitivity, specificity, and accuracy were 0.86, 88.4, 74.2, and 82.7%, respectively. Our validated folds had an average AUROC, sensitivity, specificity, and accuracy of 0.86, 80.3, 80.9, and 80.4%, respectively. Models created from the compressed videos decreased in accuracy as image sampling rate decreased from 60 Hz to 15 Hz. There was only minimal change in the accuracy of nystagmus detection when decreasing image resolution and keeping sampling rate constant.

Conclusion

Deep learning is useful in detecting nystagmus in 60 Hz video recordings as well as videos with lower image resolutions and sampling rates, making it a potentially useful tool to aid future automated eye-movement enabled neurologic diagnosis.

Screening entire health system ECG databases to identify patients at increased risk of death

BACKGROUND

Current methods to identify patients at higher risk for sudden cardiac death, primarily left ventricular ejection fraction ≤35%, miss ≈80% of patients who die suddenly. We tested the hypothesis that patients with elevated QRS-scores (index of myocardial scar) and wide QRS-T angles (index abnormal depolarization-repolarization relationship) have high 1-year all-cause mortality and could be further risk stratified with clinical characteristics.

METHODS AND RESULTS

We screened all 12-lead ECGs over 6 months at 2 large hospital systems and analyzed clinical characteristics and 1-year mortality. Patients with ECGs obtained in hospital areas with known high mortality rates were excluded. At the first hospital, QRS-score ≥5 and QRS-T angle ≥105° identified 8.0% of patients and was associated with an odds ratio of 2.79 (95% confidence interval, 2.10-3.69) for 1-year mortality compared with patients below both ECG thresholds (13.9% versus 5.5% death rate). Left ventricular ejection fraction was >35% in 82% of the former group of patients, and addition of ECG measures to left ventricular ejection fraction increased the discrimination of death risk (P<0.0001). At the second hospital, the odds ratio was 2.42 (1.95-3.01) for 1-year mortality (8.8% versus 3.8%). Adjustment for patient characteristics eliminated interhospital differences. Multivariable adjusted odds ratio combining data from both hospitals was 1.53 (1.28-1.83). Increasing heart rate and chronic renal impairment further predicted mortality.

CONCLUSIONS

Screening hospital ECG databases with QRS-scoring and QRS-T angle analysis identifies patients with high 1-year all-cause mortality and predominantly preserved left ventricular ejection fraction. This approach may represent a widely available method to identify patients at increased risk of death.

Developing computational resources in cardiac gene expression

The development of biotechnology and the completion of the Human Genome Project have led to the rapid emergence of enormous amounts of genomic information. Computational resources are needed to enable researchers and clinicians to obtain quick, up-to-date information in targeted areas. The creation of this type of subject-specific knowledge base requires expertise in diverse fields. This chapter is meant to provide a detailed overview and blueprint for how to construct a cardiac-specific knowledge base like the Human Cardiac Gene Expression Knowledge Base (CaGE) (1). We have provided an overview of some of the ways to capture, organize, and store the necessary data using Perl scripts. Each group or individual interested in developing a cardiac or gene expression knowledge base will have unique needs. We hope that this information will serve as a guide for the development of new computational tools.

CaGE: cardiac gene expression knowledgebase

CaGE is a Cardiac Gene Expression knowledgebase we have developed to facilitate the analysis of genes important to human cardiac function. CaGE integrates the functionality of the LocusLink database with data from several human cardiac expression libraries, phenotypic data from OMIM and data from large-scale microarray gene expression studies to create a knowledgebase of gene expression in human cardiac tissue. The knowledgebase is fully searchable via the web using several intuitive query interfaces. Results can be displayed in several concise easy to navigate formats.

AVAILABILITY

CaGE is located at http://www.cage.wbmei.jhu.edu

Statistical analysis of QT interval as a function of changes in RR interval in the conscious dog

INTRODUCTION

The duration of cardiac ventricular depolarization and repolarization is represented as the QT interval. QT prolongation has been associated with the occurrence of arrhythmias. Both cardiovascular as well as noncardiovascular agents have caused QT prolongation and sudden death in humans. Changes in heart rate (HR) play a major, though not exclusive, role in QT variation. Considerable debate has centered on how to normalize QT for variations in HR (QTc).

METHODS

The most common approaches use Bazett's (QTc = QT/(square root)RR) or Fridericia's (QTc = QT/(cube root)) formulas to fit the data and establish a single coefficient to analyze QT with respect to its relationship to RR, where RR= 60/HR. These single-coefficient models do not adequately describe the QT functional relationship with RR for the dog. Therefore, any calculation of QTc for the dog is misleading and can result in a false-positive indication or mask the potential hazards of a high QT. Other investigators have proposed multicoefficient exponential regression analyses to best fit the QT-RR relationship.

RESULTS AND DISCUSSION

Data presented here from dogs under resting conditions and during pharmacological maneuvers (E-4031 or cisapride intravenous infusion) support the use of such a model. In order to fully characterize drug-induced changes in the QT-RR relationship, our approach includes a statistical comparison of the regression curves for an overall effect, and quantitates the incidence and magnitude of points exceeding the upper 95% confidence interval ('outliers') to assess the degree of heterogeneity of ventricular repolarization.

Histological validation of myocardial microstructure obtained from diffusion tensor magnetic resonance imaging

Diffusion tensor magnetic resonance imaging (MRI) is a possible new means of elucidating the anatomic structure of the myocardium. It enjoys several advantages over traditional histological approaches, including the ability to rapidly measure fiber organization in isolated, perfused, arrested hearts, thereby avoiding fixation and sectioning of artifacts. However, quantitative validation of this MRI method has been lacking. Here, fiber orientations estimated in the same locations in the same heart using both diffusion tensor MRI and histology are compared in a total of two perfused rabbit hearts. Fiber orientations were statistically similar for both methods and differed on average by 12 degrees at any single location. This is similar to the 10 degrees uncertainty in fiber orientation achieved with histology. In addition, imaging studies performed in a total of seven hearts support a level of organization beyond the myofiber, the recently described laminar organization of the ventricular myocardium.

Expert systems

The concept of computerized expert systems is explained, the potential utility of these systems in pharmacy is explored, and strategies and imperatives for implementing them are described. Computerized expert systems attempt a higher level of analysis than traditional computer programs. They can be defined as systems that attempt to make or assist in a decision that is not yet completely and reliably definable in objective terms. Because of the information-intensive nature of pharmacy practice, this field is particularly suited to use of expert systems. Current applications include screening for drug interactions and therapeutic drug monitoring. Expert systems must offer a substantial advantage over human expertise (for example, by quickly analyzing enormous quantities of data); those that perform functions that humans could perform have failed to gain widespread use. An ideal hospital expert system would have access to any data available about a patient's care and would detect critical situations as they occur. Such a system would require pharmacists to shift from a prescription-based orientation to a case-management orientation. Factors to consider in implementing an expert system include linkage among multiple departments, usage options, development strategies, and maintenance requirements. Computerized expert systems hold great potential for application to pharmacy and may influence the pharmacist's role in patient care.

Personal computer-based expert system for quality assurance of antimicrobial therapy

A personal computer (PC)-based expert system developed to monitor the appropriateness of antimicrobial therapy is described. Susceptibility test data and antimicrobial therapy data are downloaded daily from the microbiology department and pharmacy department computer systems. Relational database software allows for the indexing, sorting, and manipulation necessary for analysis. The expert system accomplishes its analyses using (1) databases of organisms, antimicrobial drugs, and susceptibility cutoff values, (2) programs for evaluating pathogenicity and therapy, and (3) algorithms that determine the timing and sequence of the analysis. System output consists of discrepant therapy reports that indicate that no therapy is being given despite the presence of pathogens, that the pathogens isolated are resistant to the therapy being given, that the therapy cannot be matched with susceptibility data on the isolates, or that the therapy was discontinued too quickly. The expert system has been in continuous operation at an 800-bed referral hospital since July 1991. During an 11-month period, the system generated 1538 discrepant therapy reports. The percentage of isolates resulting in reports varied substantially with the source of the isolate. Therapy was more likely to be improved when the physician was contacted about the potential problem indicated by the report than when the physician was not contacted. A PC-based expert system using data from unlinked pharmacy and microbiology computer systems automatically evaluated the appropriateness of antimicrobial drug therapy in light of susceptibility test data.

Oxygen transport in a woman with hemoglobin Hope/beta+ thalassemia

Because their blood may "unload" oxygen more readily than normal, people with hemoglobin of low oxygen affinity might be expected to be anemic. We have studied a woman with hemoglobin Hope/beta+ thalassemia, whose hemoglobin level was 10.4 to 12.3 gm/dl (normal 14 +/- 2) despite a P50 of 41 mm Hg (normal 26). Her cardiac index was normal, yielding a calculated mixed venous PO2 of 51 mm Hg (normal 34 to 49). Oxygen transport in patients with low oxygen affinity can be maintained by a variety of homeostatic responses, only one of which is altered erythropoiesis.

Variability of the homeostatic response to altered p50

Blood from carriers of hemoglobin Osler (Hb Osler) had almost the same oxygen affinity as that of carriers of Hb McKees Rocks (Hb MR) (P50 10-11 mm Hg), but Hb concentrations were higher in male carriers of the former (21.6 versus 17.2 g/dl). Two carriers of each Hb were studied to compare their adaptations to altered oxygen affinity and their responses to phlebotomy. All four were healthy, and all excreted normal amounts of erythropoietin. Carriers of Hb MR had somewhat lower mixed venous pO2 than carriers of Hb Osler. There was no suggestion that phlebotomy impaired ability to exercise in either group of patients.

Hb Potomac (101 Glu replaced by Asp): speculations on placental oxygen transport in carriers of high-affinity hemoglobins

Blood from a woman with unexplained erythrocytosis had increased oxygen affinity, but no abnormality could be detected by electrophoresis or chromatography of her hemolysate. Separation of the tryptic peptides of her beta chains disclosed two half-sized peaks in the regions of beta T-11. The faster of these was abnormal, with the structure beta 101 Glu replaced by Asp. The new hemoglobin was called "Potomac." Three of the proband's four surviving siblings and both of her children were carriers. Differences in the ratio of carrier: normal children born to male of female carriers of 23 other high-affinity hemoglobins were not significant. The high proportion of carriers in this kindred was probably due to chance alone, and not because high maternal oxygen affinity interfered with oxygen transport to fetuses with normal hemoglobin.

Postsynthetic deamidation of hemoglobin Providence (beta 82 Lys replaced by Asn, Asp) and its effect on oxygen transport

Carriers of hemoglobin Providence have three types of beta chain in their hemolysates. The two abnormal chains have asparagine (Providence N, Prov N) or aspartic acid (Providence D) at position beta 82, instead of lysine. In vitro, only two beta chains are synthesized by reticulocytes of carriers, betaA and betaProv N. In vivo studies showed that the specific activity of Providence N was initially 10-fold higher than that of Providence D; the specific activities of the two labeled hemoglobins were approximately equal 5 wk after injection of isotope. Oxygen affinity of carriers' blood was somewhat increased, but they were not polycythemic. The affinity of the purified hemoglobins Providence was decreased. Addition of 2, 3 diphosphoglycerate had little effect on the affinity of either hemoglobin component, and addition of inositol hexaphosphate produced no change in the affinity of Providence D. These studies demonstrate that Providence N is deamidated to Providence D during the life span of the erythrocyte, and suggest this finding may represent only an easily observed prototype of posttranslational modification of proteins in general. Despite and abnormal P50 of the blood, oxygen transport is probably normal in carriers of the abnormal hemoglobins.