Breast cancer patients suffer from the symptoms of their illness as well as from burden of treatment imposed by their care. Patients with high levels of burden tend to be less compliant with treatment plans resulting in worsening outcomes. To address the problem of overburden, some providers have proposed practicing minimally disruptive medicine, where treatment plans are tailored to the patientsâ capacity to handle the care. While some researchers have developed surveys that identify and quantify factors that contribute to treatment burden, no studies have used the electronic health record to assess patient burden. We developed measures derived from outpatient and inpatient encounter data that included time spent in appointments, waiting time, unique appointment days, and total inpatient length of stay. We used these measures to differentiate burden of treatment in early stage breast cancer patients in the first eighteen months after diagnosis. This method allowed us to identify outliers and to characterize the pattern of treatment over time. Our measures could also be used to evaluate new therapeutic and operational interventions for their effect on treatment burden. In patients receiving chemotherapy at Vanderbilt, a non-inferior change in protocol successfully reduced treatment burden while a therapeutically superior treatment may have imposed an increase of burden on patients. As the complexity of healthcare increases and patients take on more responsibility to manage their care, understanding treatment burden is critical to helping providers prescribe care right-sized for the patient to improve compliance and clinical outcomes.

The widely used American Society of Anesthesiologistâsâ (ASA) Physical Status classification is subjective and requires time-consuming clinician assessment. Machine learning can be used to develop a system that predicts the ASA score a patient should be given based on routinely available preoperative data. The problem of ASA prediction is reframed into a binary classification problem for predicting between ASA 1/2 versus ASA 3/4/5. Retrospective ASA scores from the Vanderbilt Perioperative Data Warehouse are used as labels, allowing the use of supervised machine learning techniques. Routinely available preoperative data is used to select features and train four different models: logistic regression, k-nearest neighbors, random forests, and neural networks. Of the selected features, ICD9 codes were tested by incorporating temporality and hierarchy. The area under the curve (AUC) of the receiver operating characteristic (ROC) of each model on a holdout set is compared. The Cohenâs Kappa is calculated for the model versus the raw data and the model versus our anesthesiologist. Results: The best performing model was the random forest, achieving an AUC of 0.884. This model results in a 0.63 Cohenâs Kappa versus the raw data, and a 0.54 Kappa against our anesthesiologist, which is comparable to unweighted Kappa values found in literature. The results suggest that a machine learning model can predict ASA score with high AUC, and achieve agreement similar to an anesthesiologist. This demonstrates the feasibility of using this model as a standardized ASA scorer.

Cancer immunotherapies, which seek to arm the patient&rsquo;s own immune system for personalized therapy, are a promising option for effective elimination of tumors. Focused ultrasound (FUS) is one propitious method for generating anti-tumor immunotherapy, advantageous in its capacity to deliver non-ionizing, non-invasive, tumor-localized treatment; this involves the transdermal deposition of sonic energy at a focal point in the tumor, which induces acute inflammation capable of activating an anti-tumor immune response. Here, we characterize, <em>in vivo</em>, the early (48 hours) adaptive anti-tumor immune responses induced by FUS treatment of tumors. Compared to untreated tumors, tumors treated with mechanical FUS (mFUS) demonstrated increased NF-&kappa;B activation in local and distant tumors. Additionally, a &ldquo;responder&rdquo; subset of mFUS-treated mice was identified and mFUS-treated tumors exhibited an increased percent of CD4+ T cells and an increased CD4+/CD8+ T cell ratio, as compared to untreated tumors. Immunohistochemical analysis of CD4+ T cells revealed a higher presence of immunostimulatory phenotypes in mFUS-treated tumors compared to untreated tumors. Taken together, these results suggested a FUS-induced shift towards anti-tumor immune activity.

Magnetization transfer (MT) imaging has emerged as a viable alternative to conventional structural MRI indices. It has been shown to be remarkably sensitive to changes in myelin associated with pathologies such as multiple sclerosis (MS). Previous work has built a solid foundation to study the MT effect in vivo, however, the existing literature falls short of developing methods that may help provide solutions to elucidating the clinical problems associated with MS. Therefore, the overall goal of this dissertation was to further the understanding of quantitative magnetization transfer (qMT) imaging at clinical MRI field strengths to provide solutions to these clinical problems. Since qMT imaging has been shown to be sensitive to myelin pathology, these metrics were translated to areas outside of the brain, into the optic nerve and spinal cord, where radiological changes may be better correlated with clinical disability. Next, the coverage of a new MT imaging method, inhomogeneous magnetization transfer (ihMT) was expanded to cover a large 3D volume in a clinically reasonable scan time. This new acquisition strategy has been shown to be specific to WM, and thus, may provide a better indicator of changes in myelin than traditional MT imaging over a large volume. Finally, the two pool MT model was investigated to devise several different methods â one based on a new acquisition strategy, and one based on a new modeling methodology â to remove effects that confound the signal of interest in chemical exchange saturation transfer (CEST) spectra. In conclusion, qMT has been shown to be a remarkably important technique towards understanding the properties of myelin. Gaining a fundamental understanding of how myelin is affected by pathologies which affect the macromolecular structure of neural tissues may facilitate advances in the way we diagnose, treat, and hopefully cure disease. qMT may provide key contributions to this puzzle, and the studies described here have hopefully laid a foundation to drive these future discoveries.

Due to the need to acquire a series of T2-weighted images, quantitative T2 mapping protocols in magnetic resonance imaging (MRI) suffer from long scan times. In order to alleviate this problem, fast spin-echo (FSE) imaging protocols can be employed, but the resulting images contain errors in the form of smoothing and ghosting artifacts which propagate to T2 maps. This dissertation presents a new method, dubbed Multiple Echo, Caesar Cipher Acquisition and Model-Based Reconstruction (ME-CAMBREC), which explicitly accounts for k-space signal attenuation during the reconstruction step. T2 maps generated by ME-CAMBREC contained reduced artifact compared to those generated by FSE methods, while requiring only a fraction of the scan time of a multiple spin-echo protocol. For moderate-to-high acceleration factors, ME-CAMBREC outperformed parallel imaging and steady-state T2 mapping techniques. Data suitable for ME-CAMBREC can be acquired in multi-slice mode using pulse sequence interleafs, but a slice gap should be employed to limit T2 bias caused by radiofrequency profile effects. Although ME-CAMBREC can be used to generate accurate T2s in the presence of flip angle errors, it was shown that the use of an independent measure of the transmit field (B1+) will improve fitted T2 precision.

<p>A number of significant muscle diseases, such as cachexia, sarcopenia, systemic chronic inflammation, along with inflammatory myopathies share TNFα-mediated inflammation in their pathogenesis. TNFα-based inflammatory stress may directly sensitize skeletal muscle to drug-induced toxicity. The two main difficulties when assessing the role of TNFα on skeletal muscle drug toxicity using in vitro methodology are first creating an experimental model that recapitulates the basal functional responses of human skeletal muscle and second validating that the experimental model accurately represents human skeletal muscle response to TNFα. The aim of this research is to resolve these two difficulties. </p><p>We first assayed the bioenergetic profile of engineered three-dimensional human skeletal muscle myobundles to probe mitochondrial health and compared its result to muscle fatigue. Then engineered a perfusion system to measure respiration under basal conditions and electrical stimulation. Finally, to assess the effect of TNFα-induced inflammation on drug responses, we engineered 3D, human skeletal myobundles, chronically exposed them the TNFα during maturation, and measured the combined response of TNFα and the chemotherapeutic, doxorubicin, on muscle function. We concluded that the human myobundles reproduce normal muscle metabolism under both basal and maximal energy demand conditions enabling the detection of drug-induced mitochondrial toxicity. The bioenergetic health index, which is a measure of normal oxidative mitochondrial function, was inversely correlated with the extent of fatigue. The relationship between mitochondrial function and physiological muscle function suggests that the mitochondrial dysfunction produced the fatigue. The custom perfusion chamber was validated to accurately measure oxygen uptake during basal and electrical stimulation conditions. This in vitro non-invasive tool enables the investigation of human muscle physiology during exercise. Utilizing the myobundle platform, a synergistic effect of the combination of TNFα and 10nM doxorubicin was found on contractile force production. Overall, we demonstrated the feasibility of the in vitro 3D, human muscle inflammatory system to act as a drug-toxicity testing platform and validated assays to quantify drug-induced decrements in mitochondrial function.</p> / Dissertation

<p>The primary barriers to multiplexed point of care immunoassays are: (1) cost; (2) response time; and (3) sample handling. Described here is a self-contained, multiplexed immunoassay platform for point of care detection that leverages a number of enabling technologies to address these barriers. This platform is referred to as the "D4" assay, as it is composed of the following four sequential, concerted events (Figure 1): (1) Dispense (droplet of blood); (2) Dissolve (printed reagents on chip); (3) Diffuse across surface; (4) Detect binding event. </p><p>The D4 assay process begins when a finger-stick is administered and the resulting droplet of blood is applied to the surface of a detector chip. Hydrophobic ink printed onto the surface of the chip confines the blood droplet to a non-fouling region containing soluble, labile spots of detection antibodies and insoluble, non-labile spots of capture antibodies. As the soluble detection antibodies are dissolved from their printed spots by the droplet of blood, three serial events occur to generate signal (Figure 2): (1) the first half of the detection complex is formed by the binding of analytes present in blood to the stable capture agent spots; (2) diffusion of the blood laterally through the polymer brush, resulting in the dissolution and diffusion of soluble detection antibody spots; (3) solubilized detection antibodies bind to their respective analyte-capture agent spots, completing the detection complex and resulting in signal generation at the position of the non-labile capture antibody spots. </p><p>This assay relies upon the ability of labeled detection antibodies, printed into a nonfouling brush as "labile spots", to be carried by blood flow to adjacent rows of stably immobilized capture antibodies by diffusion of the analyte solution (Figure 2). Generation of signal at a given capture spot location provides identification of individual analytes (positives). Quantification of the concentration of the different analytes is carried out identically to a conventional fluorescence immunoassay by pre-calibration of the system using a dilution series of the analyte spiked into whole blood.</p><p>The D4 assay addresses several critical needs in point of care testing as follows: First, the cost of testing is reduced through miniaturization, multiplexing and one-step, on-site processing of undiluted whole blood obtained from a finger stick. Second, in order to simplify the immunoassay process, the D4 relies on diffusion to bring spatially localized reagents together to create a functional assay and thereby eliminate the need for liquid transfer steps, microfluidic manipulation of sample or reagents, and wash steps. Third, this multiplexed platform is capable of screening for a panel of markers in a single drop of blood with no sample preprocessing. Fourth, the assay is fast, which alleviates the difficulties often associated with communicating the outcome of diagnostic tests. A prototype of the D4 assay is shown in Figure 3 below.</p> / Dissertation

Cancer is a multistage, heterogeneous disease that develops through a series of genetic mutations. Early stage cancer is most responsive to treatment but can be the hardest to detect due to its small size, lack of definitive symptoms and potential location deep in the body. Whole body imaging methods, MRI/CT/PET, lack the necessary resolution to detect cellular level abnormalities. Optical methods, which have sufficient resolution, can be miniaturized into endoscopes, which are necessary to overcome limited penetration of light into tissue. By combining optical coherence tomography (OCT) and fluorescence imaging methods it is possible to create endoscopes sensitive to molecular and structural changes. I applied a dual-modality 2mm diameter rigid endoscope to the study of the natural history of colon cancer in a mouse model, and later applied this knowledge to the design and characterization of a 0.8 mm dual-modality flexible probe for use in human fallopian tubes. By using this endoscope, which is introduced through the natural orifice and is compatible with existing hysteroscopes, high-risk women could be screened in a procedure at a similar level of invasiveness as a colonoscopy. Therefore, the endoscope fills this gap in clinical care for women at high-risk for ovarian cancer.

<p> Kinesthesia refers to sensations of limb position and movement, and deficits of upper limb kinesthetic feedback are common after stroke, impairing stroke survivors&rsquo; ability to perform the fundamental reaching and stabilization behaviors needed for daily functions like self-feeding. I attempt to mitigate the negative impact of post-stroke kinesthesia deficits by evaluating the utility of vibrotactile sensory substitution to restore closed-loop kinesthetic feedback of the upper limb.</p><p> As a first step, this study evaluated performance in healthy individuals during fundamental reaching, stabilization, and tracking behaviors while using supplemental vibrotactile feedback encoding either limb state information or goal-aware error information. First, I determined that performance in reaching and stabilization tasks varies systematically with the amount of limb position and velocity information encoded in limb state feedback and that there is an optimal combination. Next, I compared the utility of optimal limb state to goal-aware error feedback. Both types of feedback reduced error in the reaching and stabilization tasks. Random task-irrelevant sham feedback did not reduce error, demonstrating participants could perceive and understand the information contained within the vibrotactile feedback. Error feedback improved performance more than state feedback; however the relative difficulty of using error feedback outside of a laboratory setting means state feedback should not be discounted. The performance while tracking could not be quantified due to issues with the task design.</p><p> As a second step, I performed a series of case studies in five chronic stroke survivors. The stroke survivors all tolerated the vibrotactile feedback well and were able to perceive and understand at least one of the limb state or error feedback encodings. Stroke survivors practiced each information encoding type for one session. During this short period our stroke survivors struggled to integrate visual and vibrotactile inputs and motor control in order to use the vibrotactile information to control the arm. However, two additional practice sessions with error feedback for one participant led to a two thirds reduction in reaching error. These results suggest stroke survivors can learn to use supplemental vibrotactile feedback to enhance control of the contralesional arm.</p>

<p>Cortical representations of rhythmic and discrete movements are analyzed and used to create a novel neural decoding algorithm for brain-machine interfaces. This algorithm is then implemented to decode both cyclic movements and reach-and-hold movements in awake behaving rhesus macaques using their cortical activity alone. Finally, a healthy macaque wears and controls a lower body exoskeleton using the developed BMIas a proof of concept of a brain-controlled neuroprosthetic device for locomotion.</p> / Dissertation