Blood analysis is crucial for early cancer screening and improving patient survival rates. However, developing an effective strategy for early cancer detection using high-throughput blood analysis is still challenging. Herein, a novel automatic super-hydrophobic platform is developed together with a deep learning (DL)-based label-free serum and surface-enhanced Raman scattering (SERS), along with an automatic high-throughput Raman spectrometer to build an effective point-of-care diagnosis system. A total of 695 high-quality serum SERS spectra are obtained from 203 healthy volunteers, 77 leukemia M5, 94 hepatitis B virus, and 321 breast cancer patients. Serum SERS signals from the normal (n = 183) and patient (n = 443) groups are used to assess the DL model, which classify them with a maximum accuracy of 100%. Furthermore, when SERS is combined with DL, it exhibits excellent diagnostic accuracy (98.6%) for the external held-out test set, indicating that this method can be used to develop a high throughput, rapid, and label-free tool for screening diseases.

No abstract

An innovative technique for the synthesis of a nonuniform sonar array with as few elements as possible is proposed. This paper introduces a new method for linear array pattern synthesis based on continuous compressive sensing (CCS). Given a reference pattern subject to aperture constraints, the proposed approach allows the exact solution of the positions of sonar sensors, whose number is maximally reduced. The excitations and positions of sensors are exactly estimated without constraints of sampling grids used in convetional CS methods. Numerical experiments show that the proposed synthesis method is able to maximally save array elements for the design of sparse linear arrays, also compared to state-of-the-art sparse array synthesis methods.Index Terms-Array design; pattern synthesis; sparse nonuniform linear array; continuous compressive sensing.978-1-4673-9724-7/16/$31.00 ©2016 IEEE