Interpretable Classification of Bacterial Raman Spectra With Knockoff Wavelets.

Deep neural networks and other machine learning models are widely applied to biomedical signal data because they can detect complex patterns and compute accurate predictions. However, the difficulty of interpreting such models is a limitation, especially for applications involving high-stakes decision, including the identification of bacterial infections. This paper considers fast Raman spectroscopy data and demonstrates that a logistic regression model with carefully selected features achieves accuracy comparable to that of neural networks, while being much simpler and more transparent. Our analysis leverages wavelet features with intuitive chemical interpretations, and performs controlled variable selection with knockoffs to ensure the predictors are relevant and non-redundant. Although we focus on a particular data set, the proposed approach is broadly applicable to other types of signal data for which interpretability may be important.

ALD HfO2 Films for Defining Microelectrodes for Electrochemical Sensing and Other Applications.

Microelectrodes are used in a wide range of applications from analytical electrochemistry and biomolecular sensing to in vivo implants. While a variety of insulating materials have been used to define the microelectrode active area, most are not suitable for nanoscale electrodes (<1 µm2) due to the limited robustness of these films when the film thickness is on the order of the nanoelectrode dimension. In this study, we investigate atomic layer deposited hafnium dioxide (ALD HfO2) as an insulating film to coat planar platinum microelectrodes, with the active areas being defined where the HfO2 is etched. Thermally grown films with thicknesses between 10 and 60 nm were deposited by 100 to 550 ALD cycles and were initially characterized by measuring their standard electrical properties and imaging incipient texture development. Electrochemical measurements on the structures were made, including linear sweep voltammetry and electrochemical impedance spectroscopy, which identified the presence of pinholes in films deposited over the range of 100 to 350 cycles, resulting in leakage. These measurements also suggest a lower limit to the size of microelectrodes below which the electrochemical current detected is no longer dominated by that through the exposed active area. A bilayer insulator comprising ALD HfO2 coated with parylene-C was investigated to minimize the pinhole leakage. Steady-state currents were measured for different electrode areas, qualitatively agreeing with the theory for areas down to ∼1 µm2. For sub-square micrometer electrode areas, bilayer-insulated devices with parylene-C apertures that exposed the smallest microelectrode area showed measured currents that were consistent with extrapolations, indicating that it reduces leakage through HfO2.

A proton NMR study on the hydration of normal versus psoriatic stratum corneum: linking distinguishable reservoirs to anatomical structures.

The NMR behaviour of normal and psoriatic stratum corneum (SC) was investigated as a function of hydration with the aim of obtaining a better understanding of the role of water in the SC structure. Time domain NMR techniques were employed to identify the signal from water and that from nonaqueous components of the SC, such as lipids and proteins. The signals were investigated as a function of water content. The free induction decay was separated into mobile signal (from water and mobile lipids) and solid signal (from protein and 'solid' lipids). Spin-spin relaxation (T(2)) measurements further separated the mobile domains within the SC. The results suggested that, when water is added to dry SC, it first enters the corneocytes; then, at a hydration of 0.24-0.33 g H(2)O/g SC (normal SC) or 0.12-0.24 g H(2)O/g SC (psoriatic SC), water begins to accumulate in hydrated lipid regions. Water was found to exchange between these two domains on the time scale of a few hundred milliseconds. When compared with normal SC, psoriatic SC had a looser corneocyte structure, a larger mobile lipid component at low hydration and a smaller capacity for corneocyte water.

Reproducibility of in vivo magnetic resonance imaging-based measurement of myelin water.

PURPOSE: To evaluate the reproducibility of multicomponent quantitative T(2) (QT2) measurements, in particular the myelin water fraction (MWF), to determine the sensitivity of this method for monitoring myelin changes in longitudinal studies and to provide a basis for correctly powering such studies. MATERIALS AND METHODS: The de facto standard 32-echo spin-echo imaging sequence was used throughout, and data were analyzed using regularized non-negative least squares (NNLS) to produce T(2) distributions. Three studies were conducted in healthy subjects. First, two acquisition protocols were compared in 10 subjects. Second, variability of QT2 was evaluated over same-day scan-rescan experiments in 6 subjects. Finally, variability was quantified in a longitudinal study of 5 subjects. RESULTS: A within-subject coefficient of variation (CoV) of 12% (range 4-25%) was observed for the MWF in brain white matter (WM) regions of interest (ROIs). The geometric mean T(2) was more stable, with a longitudinal CoV of 4% (range 1-6%). The choice of the geometry and repetition time of the acquisition protocol influenced the estimates of the MWF and T(2) values. The choice of integration range for the short-T(2) component had a significant effect on MWF estimates, but not on reproducibility. CONCLUSION: The reproducibility of QT2 measurements using existing methods is moderate and the method can be used in longitudinal studies, with careful consideration of the methodologic variability and an appropriate group size.