Rui Costa Martins

Abstract

Abstract
Spectral information is characterized by multi-scaled interference, convolution and variability. Spectral lines are fragmented and diffused along the spectra. In many cases, matrix and physical effects do not allow to determine specific bands. Despite this limitation, the observed spectra contains significant amounts of information about the sample composition and characteristics, which once understood, can make spectroscopy an ideal technology for analyzing complex samples, such as bodyfluids and tissues. Breaking down and deciphering the structure of spectral information is paramount for the development of reagent-free point-of-care devices. A self-learning artificial intelligence was developed to take advantage of spectral complex information structure. It determines the relationships between composition and/or spectral features in high-dimensional space, where different sub-spaces correlate to specific constituents or characteristics. It also establishes a knowledgebase, by feature space transformations and optimizing co-variance search direction under the correct 'matrix effect' context. An example of hemogram analysis with erythrocyte and leucocyte counts is presented to demonstrate the advantages of the developed methodology.

Abstract
Tuberculosis is one of the major public health concerns. This highly contagious disease affects more than 10.4 million people, being a leading cause of morbidity by infection. Tuberculosis is diagnosed at the point-of-care by the Ziehl-Neelsen sputum smear microscopy test. Ziehl-Neelsen is laborious, prone to human error and infection risk, with a limit of detection of 104 cells/mL. In resource-poor nations, a more practical test, with lower detection limit, is paramount. This work uses a magnetoresistive biosensor to detect BCG bacteria for tuberculosis diagnosis. Herein we report: i) nanoparticle assembly method and specificity for tuberculosis detection; ii) demonstration of proportionality between BCG cell concentration and magnetoresistive voltage signal; iii) application of multiplicative signal correction for systematic effects removal; iv) investigation of calibration effectiveness using chemometrics methods; and v) comparison with state-of-the-art point-of-care tuberculosis biosensors. Results present a clear correspondence between voltage signal and cell concentration. Multiplicative signal correction removes baseline shifts within and between biochip sensors, allowing accurate and precise voltage signal between different biochips. The corrected signal was used for multivariate regression models, which significantly decreased the calibration standard error from 0.50 to 0.03 log10 (cells/mL). Results show that Ziehl-Neelsen detection limits and below are achievable with the magnetoresistive biochip, when pre-processing and chemometrics are used. © 2017 Elsevier B.V.

Abstract

Abstract
OpenCL is an open standard for parallel programming of heterogeneous compute devices, such as GPUs, CPUs, DSPs or FPGAs. However, the verbosity of its C host API can hinder application development. In this paper we present cf4ocl, a software library for rapid development of OpenCL programs in pure C. It aims to reduce the verbosity of the OpenCL API, offering straightforward memory management, integrated profiling of events (e.g., kernel execution and data transfers), simple but extensible device selection mechanism and user-friendly error management. We compare two versions of a conceptual application example, one based on cf4ocl, the other developed directly with the OpenCL host API. Results show that the former is simpler to implement and offers more features, at the cost of an effectively negligible computational overhead. Additionally, the tools provided with cf4ocl allowed for a quick analysis on how to optimize the application.