Data-driven development of sparse multi-spectral sensors for urological tissue differentiation

Abstract

Infrared spectroscopy is often used to spot differences between benign and malignant tissue. Due to the proliferation of tumorous cells, the composition of tissue changes drastically. In the consequence shifts occur in its optical properties that are indicated by spectral biomarkers in the so-called fingerprint region. In this work, we propose a new concept for a sparsified multi-spectral measurement of the most important and informative biomarker signals. The results of a data-driven feature selection approach show that a reliable discrimination of the tissue is still possible, even though utilizing only a small fraction of the measured data. A selected arrangement of only a few narrow-band quantum cascade lasers could provide proficient signal-to-noise ratios and can noticeably reduce the data acquisition time. Consequentially, real-time applications will be possible in short-term and in-vivo diagnostics in the long-term. First measurements of silicone phantoms validate the imaging capability of the sensor concept.

Keywords

• biomarker detection
• feature selection
• infrared spectroscopy
• tissue differentiation