Abstract 862: A method to create aggregate histogram representations for machine learning from small sample size medical sensor time series

Background: Machine learning has been used to successfully develop biomarkers from data collected from a wide range of commodity and specialized sensor devices. Typical machine learning algorithms, such as deep learning, require several thousand to on the order of millions of samples of data to work most effectively. However, in many studies with limited numbers of patients and data samples, such methods cannot be used.

Objective: To solve this problem, we have developed a specialized algorithm that can convert time series signals into a set of histogram signatures that can then be used with unsupervised methods such as clustering to create biomarkers.

Methods: Our methods decompose the time series data into time windows and analyze aggregate mean and variance and other statistical qualities over these windows.

Results: We will describe our approach and show how we have applied it along with genetic algorithms to perform auto-tuning for novel biomarkers and endpoints.

Conclusions: We will be able to develop a highly accurate biomarker to detect gate imbalance and impairment from limited accelerometer data.

Citation Format: Nick Moss, Ingrid Oakley-Girvan, Michelle R. Longmire. A method to create aggregate histogram representations for machine learning from small sample size medical sensor time series [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 862.

Abstract 2033: Reducing cancer caregiver burden: A user-centered design approach for an mHealth app

Background: Informal cancer caregivers are being asked to provide more extensive home care for cancer patients. It is well documented that caregivers frequently experience psychological and behavioral effects that impact their overall well-being. Digital solutions can support informal caregivers by providing key information on the patient's care and nudging them to focus on their own mental and physical health.

Objective: In this paper, we describe a user-centered design approach to build a digital solution by engaging with stakeholders that include clinical staff, health care decision makers, cancer patients and survivors, and informal caregivers. The NIH funded digital solution, TOGETHERCareTM, (Track Outcomes & Guidance, Technology for Health & Effective Resources for Care) is a smart phone mobile app to provide support and resources to informal caregivers while enabling them to monitor the cancer survivor's health for potential adverse events, thereby preventing unnecessary emergency care and reducing daily caregiver worries.

Methods: An iterative information gathering process was conducted that included a) key-informant (n=139) and semi-structured interviews (n=11) with stakeholders to assess health care value propositions and corresponding modules of benefit; b) prototype review, evaluation and feedback by a Patient Advisory Council and Clinical Advisory Team and; c) a 28-day beta user-testing with feasibility and acceptability feedback from 15 stakeholders in two geographically diverse academic cancer centers (Duke and Stanford). This study was registered on clinicaltrials.gov (NCT04018677).

Results: The interviews conducted prior to developing the mHealth app prototype identified consistency in responses between different stakeholder groups in terms of how the mobile app should work, and areas of difference. Clinical teams focused on app usage efficiency and features that would improve information sharing and follow-up visits with survivors. Survivors and caregivers were more focused on features that would provide assistance with at-home medical tasks and activities of daily living. The interviews and data analysis following the Beta test of the prototype indicated satisfaction with the app's usability, that caregivers preferred to focus primarily on the patient's health and not their own, and that regular surveys on the patient's symptoms educated and reduced anxiety among caregivers.

Conclusions: This study describes the user-centered design process and demonstrates the feasibility and acceptability of TOGETHERCareTM, an iOS smart phone app for informal cancer caregivers. Larger studies, in various oncology populations, are needed to further evaluate efficacy of the app across health metrics that matter to the different stakeholders.

Citation Format: Ingrid Oakley-Girvan, Vasu Divi, Oxana Palesh, Jena Daniels, Lisa G. Rosas, Dale O'Brien, Sharon W. Davis, Arif H. Kamal, Allison W. Kurian, Michelle R. Longmire. Reducing cancer caregiver burden: A user-centered design approach for an mHealth app [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2033.

A pilot study on providing ophthalmic training to medical students while initiating a sustainable eye care effort for the underserved

IMPORTANCE

We present a method to reintroduce ophthalmic training into the medical school curriculum.

OBJECTIVES

To evaluate knowledge and skills acquired when participating in a service project, the Community Vision Project, and to develop a quantitative method for testing skills with the direct ophthalmoscope in patients.

DESIGN

Second-year medical students participated in the study. After 1 month, their knowledge was compared with that of peers and graduates (internal medicine residents). Also at 1 month, their direct ophthalmoscope skills were compared with those of upperclassmen who had completed all core clerkships. One year later, after the participants had completed their core clerkships, long-term ophthalmoscope skills retention was tested, and their performance was compared with that of their classmates.

SETTING AND PARTICIPANTS

Training occurred in mobile eye clinics. Knowledge and skills assessments were performed in the hospital eye clinic among students and residents at The University of New Mexico School of Medicine. Patients were recruited from the hospital eye clinic. Participants attended a 3-hour training session held by an attending physician in the hospital eye clinic and took part in at least 1 mobile eye clinic.

MAIN OUTCOMES AND MEASURES

A knowledge assessment quiz was administered to participants (n = 12), their classmates (n = 18), and internal medicine residents (n = 33). Skills assessment with the direct ophthalmoscope was performed at 1 month and at 1 year in 5 participants and 5 nonparticipants. Tonometer skills were assessed by comparing participants' readings with those of an ophthalmologist's obtained in patients at the mobile eye clinics. RESULTS Participants' median knowledge assessment scores were 48% higher than those of their classmates and 37% higher than those of internal medicine residents (P < .001 for both). Short-term (1 month) direct ophthalmoscopy median scores were 60% (quartile 1 to quartile 3 range, 40%-80%) for participants and 40% (quartile 1 to quartile 3 range, 20%-60%) for nonparticipating upperclassmen (P = .24). Long-term direct ophthalmoscopy median scores were 100% (quartile 1 to quartile 3 range, 75%-100%) for participants and 0% (quartile 1 to quartile 3 range, 0%-25%) for nonparticipating classmates (P = .11). Participants' tonometer readings were similar to those of the ophthalmologist's; their median reading was 2 mm Hg (quartile 1 to quartile 3 range, 0-4 mm Hg) higher than that of the ophthalmologist's (P = .05, sign test).

CONCLUSIONS AND RELEVANCE

Service-based learning offered an efficient model for incorporating ophthalmic training into the medical school curriculum. A viable tool for quantitatively testing ophthalmoscope skills is presented.

Dendrimers as high relaxivity MR contrast agents

Dendrimers are versatile macromolecules with tremendous potential as magnetic resonance imaging (MRI) contrast agents. Dendrimer-based agents provide distinct advantages over low-molecular-weight gadolinium chelates, including enhanced r1 relaxivity due to slow rotational dynamics, tunable pharmacokinetics that can be adapted for blood pool, liver, kidney, and lymphatic imaging, the ability to be a drug carrier, and flexibility for labeling due to their inherent multivalency. Clinical applications are increasingly being developed, particularly in lymphatic imaging. Herein we present a broad overview of dendrimer-based MRI contrast agents with attention to the unique chemistry and physical properties as well as emerging clinical applications.

Polychromatic in vivo imaging of multiple targets using visible and near infrared light

Conventional diagnostic imaging methods such as X-ray CT, MRI, and nuclear medicine are inherently monochromatic meaning that they can depict only one molecular target at a time. Optical imaging has the unique ability to be polychromatic and therefore multi-color imaging employing targeted agents conjugated to fluorophores of varying wavelength enables multiple simultaneous readouts thus providing greater multiplexed information. Numerous successful multicolor imaging techniques have recently been reported using optical imaging in in vivo animal disease models, thus adding to a growing body of research supporting the clinical viability and applicability of these technologies. Herein, we review multicolor optical imaging from the basic chemistry and physics perspective and then extend this to biological and medical applications.

Recent advances in optical cancer imaging of EGF receptors

Epidermal growth factor (EGF) receptors are commonly expressed on the cell membrane of cancer cells and activity of these receptors results in accelerated cell growth and carcinogenesis. A variety of targeted molecules have been developed to block ligand binding and/or inhibit the function of these receptor tyrosine kinases, and several have proven therapeutic benefits. Along with the advent of new therapeutic agents comes a need for non-invasive tools to diagnose, characterize, and monitor tumor responsiveness to therapy. Imaging EGF receptors with radionuclides has been performed for decades. However, recently this area has advanced considerably with the development of EGF receptor-targeted optical imaging probes. Herein, we review recent advances in molecular imaging of the EGF receptor family, focusing specifically on optical imaging. Such agents provide the opportunity for earlier diagnosis, improved tumor characterization, and the ability to measure and monitor tumor responsiveness to anti-EGF receptor treatment strategies.

Near infrared fluorescence-guided real-time endoscopic detection of peritoneal ovarian cancer nodules using intravenously injected indocyanine green

Near infrared fluorescence-guidance can be used for the detection of small cancer metastases and can aid in the endoscopic management of cancer. Indocyanine green (ICG) is a Food and Drug Administration (FDA)-approved fluorescence agent. Through non-specific interactions with serum proteins, ICG achieves enhanced permeability and retention (EPR) effects. Yet, ICG demonstrates rapid clearance from the circulation. Therefore, ICG may be an ideal contrast agent for real-time fluorescence imaging of tumors. To evaluate the usefulness of real-time dual fluorescence and white light endoscopic optical imaging to detect tumor implants using the contrast agent ICG, fluorescence-guided laparoscopic procedures were performed in mouse models of peritoneally disseminated ovarian cancers. Animals were administered intravenous ICG or a control contrast agent, IR800-conjugated to albumin. The ability to detect small ovarian cancer implants was then compared. Using the dual view microendoscope, ICG clearly enabled visualization of peritoneal ovarian cancer metastatic nodules derived from SHIN3 and OVCAR5 cells at 6 and 24 hr after injection with significantly higher tumor-to-background ratio than the control agent, IR800-albumin (p < 0.001). In conclusion, ICG has the desirable properties of having both EPR effects and rapid clearance for the real-time endoscopic detection of tiny ovarian cancer peritoneal implants compared to a control macromolecular agent with theoretically better EPR effects but longer circulatory retention. Given that ICG is already FDA-approved and has a long track record of human use, this method could be easily translated to the clinic as a robust tool for fluorescence-guided endoscopic procedures for the management and treatment of cancer.

Rational chemical design of the next generation of molecular imaging probes based on physics and biology: mixing modalities, colors and signals

In recent years, numerous in vivo molecular imaging probes have been developed. As a consequence, much has been published on the design and synthesis of molecular imaging probes focusing on each modality, each type of material, or each target disease. More recently, second generation molecular imaging probes with unique, multi-functional, or multiplexed characteristics have been designed. This critical review focuses on (i) molecular imaging using combinations of modalities and signals that employ the full range of the electromagnetic spectra, (ii) optimized chemical design of molecular imaging probes for in vivo kinetics based on biology and physiology across a range of physical sizes, (iii) practical examples of second generation molecular imaging probes designed to extract complementary data from targets using multiple modalities, color, and comprehensive signals (277 references).

Biologically optimized nanosized molecules and particles: more than just size

The expanded biological and medical applications of nanomaterials place a premium on better understanding of the chemical and physical determinants of in vivo particles. Nanotechnology allows us to design a vast array of molecules with distinct chemical and biological characteristics, each with a specific size, charge, hydrophilicity, shape, and flexibility. To date, much research has focused on the role of particle size as a determinant of biodistribution and clearance. Additionally, much of what we know about the relationship between nanoparticle traits and pharmacokinetics has involved research limited to the gross average hydrodynamic size. Yet, other features such as particle shape and flexibility affect in vivo behavior and become increasingly important for designing and synthesizing nanosized molecules. Herein, we discuss determinants of in vivo behavior of nanosized molecules used as imaging agents with a focus on dendrimer-based contrast agents. We aim to discuss often overlooked or, yet to be considered, factors that affect in vivo behavior of synthetic nanosized molecules, as well as aim to highlight important gaps in current understanding.

Multiplexed imaging in cancer diagnosis: applications and future advances

The development of imaging technologies that have sufficient specificity and sensitivity to enable early, accurate detection of cancer and response to therapy has long been a goal in oncology. Various radiological techniques have been used for diagnosis and surveillance of disease recurrence and imaging has revolutionised oncology. However, despite the widespread use of technologies, the ability of currently available imaging methods to facilitate early detection, precise characterisation, and accurate localisation of malignant disease could be improved. The simultaneous use of two or more techniques, contrast reagents, signalling methods, or the coupling of agent and tissue properties to achieve so-called multiplexed imaging is a promising approach. In this review, we provide a broad overview of current and emerging multiplexed, imaging technologies.

Fluorophore-quencher based activatable targeted optical probes for detecting in vivo cancer metastases

In vivo molecularly targeted fluorescence imaging of tumors has been proposed as a strategy for improving cancer detection and management. Activatable fluorophores, which increased their fluorescence by 10-fold after binding tumor cells, result in much higher target to background ratios than conventional fluorophores. We developed an in vivo targeted activatable optical imaging probe based on a fluorophore-quencher pair, bound to a targeting moiety. With this system, fluorescence is quenched by the fluorophore-quencher interaction outside cancer cells, but is activated within the target cells by dissociation of the fluorophore-quencher pair. We selected the TAMRA (fluorophore)-QSY7 (quencher) pair and conjugated it to either avidin (targeting the D-galactose receptor) or trastuzumab (a monoclonal antibody against the human epithelial growth factor receptor type2 (HER2/neu)) and evaluated their performance in mouse models of cancer. Two probes, TAMRA-QSY7 conjugated avidin (Av-TM-Q7) and trastuzumab (Traz-TM-Q7) were synthesized. Both demonstrated better than similar self-quenching probes. In vitro fluorescence microscopic studies of SHIN3 and NIH/3T3/HER2+ cells demonstrated that Av-TM-Q7 and Traz-TM-Q7 produced high intracellular fluorescent signal. In vivo imaging with Av-TM-Q7 and Traz-TM-Q7 in mice enabled the detection of small tumors. This molecular imaging probe, based on a fluorophore-quencher pair conjugated to a targeting ligand, successfully detected tumors in vivo due to its high activation ratio and low background signal. Thus, these activatable probes, based on the fluorophore-quencher system, hold promise clinically for "see and treat" strategies of cancer management.

Multi-targeted multi-color in vivo optical imaging in a model of disseminated peritoneal ovarian cancer

Commonly used in flow cytometry, multiplexed optical probes can diagnose multiple types of cell surface marker, potentially leading to improved diagnosis accuracy in vivo. Herein, we demonstrate the targeting of two different tumor markers in models of disseminated ovarian cancer. Two ovarian cancer cell lines (SKOV3 and SHIN3) were employed; both overexpress D-galactose receptor (D-galR), but only SKOV3 overexpresses HER2/neu. Additionally, fusion tumors composed of SKOV3 and SHIN3/RFP were evaluated. Both galactosyl serum albumin-rhodamine green (GSA-RhodG), which binds D-galR, and trastuzumab-Alexa680, which binds HER2/neu, were administered to tumor-bearing mice for in vivo fluorescence imaging and in situ fluorescence microscopy. In vivo fluorescence imaging depicted 64 of 69 SKOV3 tumors (94.2%) based on their dual spectra corresponding to both RhodG and Alexa680, while all 71 SHIN3 tumors (100%) were detected based on their single spectrum corresponding only to RhodG. All 59 SHIN3 and 36 SKOV3 tumors were correctly diagnosed with in situ microscopy. Additionally, in the mixed tumor model, all tumors could be depicted using the RhodG spectrum, but only SKOV3 components also showed the Alexa680 spectrum. In conclusion, multitargeted multicolor optical imaging enabled specific in vivo diagnosis of tumors expressing distinct patterns of receptors, leading to improved diagnostic accuracy.

Determination of optimal rhodamine fluorophore for in vivo optical imaging

Optical imaging has the potential to improve the efficacy of surgical and endoscopic approaches to cancer treatment; however, the optimal type of fluorescent probe has not yet been established. It is well-known that rhodamine-core-derived fluorophores offer a combination of desirable properties such as good photostability, high extinction coefficient, and high fluorescence quantum yield. However, despite the ubiquitous use of rhodamine fluorophores for in vivo optical imaging, it remains to be determined if unique chemical properties among individual rhodamine core family members affect fluorophore parameters critical to in vivo optical imaging applications. These parameters include preserved fluorescence intensity in low pH environments, similar to that of the endolysosome; efficient fluorescence signal despite conformational changes to targeting proteins as may occur in harsh subcellular environments; persistence of fluorescence after cellular internalization; and sufficient signal-to-background ratios to permit the identification of fluorophore-targeted tumors. In the present study, we conjugated 4 common rhodamine-core based fluorescent dyes to a clinically feasible and quickly internalizing D-galactose receptor targeting reagent, galactosamine serum albumin (GmSA), and conducted a series of in vitro and in vivo experiments using a metastatic ovarian cancer mouse model to determine if differences in optical imaging properties exist among rhodamine fluorophores and if so, which rhodamine core possesses optimal characteristics for in vivo imaging applications. Herein, we demonstrate that the rhodamine-fluorophore, TAMRA, is the most robust of the 4 common rhodamine fluorophores for in vivo optical imaging of ovarian cancer metastases to the peritoneum.