The current methods for diagnosis of acute and chronic infections are complex and skill-intensive. For complex clinical biofilm infections, it can take days from collecting and processing a patient’s sample to achieving a result. These aspects place a significant burden on healthcare providers, delay treatment, and can lead to adverse patient outcomes. We report the development and application of a novel multi-excitation Raman spectroscopy-based methodology for the label-free and non-invasive detection of microbial pathogens that can be used with unprocessed clinical samples directly and provide rapid data to inform diagnosis by a medical professional. The method relies on the differential excitation of non-resonant and resonant molecular components in bacterial cells to enhance the molecular finger-printing capability to obtain strain-level distinction in bacterial species. Here, we use this strategy to detect and characterize the respiratory pathogens Pseudomonas aeruginosa and Staphylococcus aureus as typical infectious agents associated with cystic fibrosis. Planktonic specimens were analyzed both in isolation and in artificial sputum media. The resonance Raman components, excited at different wavelengths, were characterized as carotenoids and porphyrins. By combining the more informative multi-excitation Raman spectra with multivariate analysis (support vector machine) the accuracy was found to be 99.75% for both species (across all strains), including 100% accuracy for drug-sensitive and drug-resistant S. aureus. The results demonstrate that our methodology based on multi-excitation Raman spectroscopy can underpin the development of a powerful platform for the rapid and reagentless detection of clinical pathogens to support diagnosis by a medical expert, in this case relevant to cystic fibrosis. Such a platform could provide translatable diagnostic solutions in a variety of disease areas and also be utilized for the rapid detection of anti-microbial resistance.

Current methods for diagnosing acute and complex infections mostly rely on culture-based methods and, for biofilms, fluorescence in-situ hybridization. These techniques are labor-intensive and can take 2-4 days to return a test result, especially considering an extra culturing step required for the antibiotic susceptibility testing (AST). This places a significant burden on healthcare providers, delaying treatment and leading to adverse patient outcomes. Here, we report the complementary use of our newly developed multi-excitation Raman spectroscopy (ME-RS) method with whole-genome sequencing (WGS). Four WHO priority pathogens are AST phenotyped and their antimicrobial resistance (AMR) profile determined by WGS. On application of ME-RS method we find high correlation with the WGS characterization. Highly accurate classification based on the species (98.93%), wild-type/non-wild type (99.45%), and presence or absence of thick peptidoglycan layers in cell walls (100%), as well as at the individual strain level (99.29%). These results clearly demonstrate the potential of ME-RS as a rapid and first-stage tool for species, resistance and strain-level classification which can be followed up by WGS for confirmation. Such a workflow can facilitate efficient antimicrobial stewardship to handle and prevent the spread of AMR.

We report a 5.8-W deep-ultraviolet (DUV) laser obtained from frequency-quadrupling of an all-fiberized ytterbium-doped fiber (YDF) master oscillator power amplifier (MOPA). The MOPA system delivers 585 ps pulses at 1040 nm with a maximum available output power of 23.5 W for nonlinear frequency conversion. A lithium triborate (LBO) crystal and a beta barium borate (BBO) crystal are employed for second- and fourth-harmonic generation (FHG), respectively. At a repetition rate of 1.6 MHz, a maximum DUV output power of 5.8 W is obtained at 260 nm with a corresponding pulse energy of 3.6 μJ and maximum peak power of at least 6.9 kW. A 1μm-to-260nm conversion efficiency of 26.4% is achieved at a DUV output power of 5.8 W. To the best of our knowledge these results represent the highest-average-power fiberized-laser-pumped DUV laser, as well as the most efficient DUV generation based on BBO crystals to date. We further demonstrate application of the pulsed DUV laser in bacterial disinfection achieving an inactivation efficiency of 99.999% for E-coli bacteria at a DUV exposure of 7 mJ/cm2.

Light sheet microscopy (LSM) has emerged as one of most profound three dimensional (3D) imaging tools in the life sciences over the last decade. However, LSM is currently performed with fluorescence detection on one- or multi-photon excitation. Label-free LSM imaging approaches have been rather limited. Second Harmonic Generation (SHG) imaging is a label-free technique that has enabled detailed investigation of collagenous structures, including its distribution and remodelling in cancers and respiratory tissue, and how these link to disease. SHG is generally regarded as having only forward- and back-scattering components, apparently precluding the orthogonal detection geometry used in Light Sheet Microscopy. In this work we demonstrate SHG imaging on a light sheet microscope (SHG-LSM) using a rotated Airy beam configuration that demonstrates a powerful new approach to direct (without any further processing or deconvolution) 3D imaging of harmonophores such as collagen fibres in biological samples. We provide unambiguous identification of SHG signals on the LSM through its wavelength and polarisation sensitivity. In a multimodal LSM setup we demonstrate that SHG and two-photon signals can be acquired on multiple types of different biological samples. We further show that SHG-LSM is sensitive to changes in collagen synthesis within lung fibroblast 3D cell cultures. This work expands on the existing optical methods available for use with light sheet microscopy, adding a further label-free imaging technique which can be combined with other detection modalities to realise a powerful multi-modal microscope for 3D bioimaging.

State-of-art NPUs are typically architected as a self-contained sub-system with multiple heterogeneous hardware computing modules, and a dataflow-driven programming model. There lacks well-established methodology and tools in the industry to evaluate and compare the performance of NPUs from different architectures. We present an event-based performance modeling framework, VPU-EM, targeting scalable performance evaluation of modern NPUs across diversified AI workloads. The framework adopts high-level event-based system-simulation methodology to abstract away design details for speed, while maintaining hardware pipelining, concurrency and interaction with software task scheduling. It is natively developed in Python and built to interface directly with AI frameworks such as Tensorflow, PyTorch, ONNX and OpenVINO, linking various in-house NPU graph compilers to achieve optimized full model performance. Furthermore, VPU-EM also provides the capability to model power characteristics of NPU in Power-EM mode to enable joint performance/power analysis. Using VPU-EM, we conduct performance/power analysis of models from representative neural network architecture. We demonstrate that even though this framework is developed for Intel VPU, an Intel in-house NPU IP technology, the methodology can be generalized for analysis of modern NPUs.