Classification of urinary stones using near-infrared spectroscopy and chemometrics: A promising method for intraoperative application

Abstract

In low-invasive surgical treatment of urolithiasis, there is a need for an analytical method to determine the chemical composition of urinary stones in real-time mode, i.e., intraoperatively. While a thorough phase analysis can be done after the surgery, preliminary information about a target stone would be helpful for the specialists for choosing an optimal strategy of treatment and giving some immediate dietary or drug prescriptions to a patient. Near-infrared spectroscopy (NIRS) is a good candidate for such a method that can provide immediate results without obligatory sample preparation. Fiber optic probes, often used for acquiring near-infrared spectra, are compatible with surgical instrumentation. Chemometric algorithms can successfully resolve the complexity of NIR spectra, which consist of overlapped signals. For the first time, we applied NIRS in diffuse reflectance mode to classify three major types of urinary stones: oxalates, urates, and phosphates. To imitate the real conditions of a surgery, the NIR spectra were acquired not only under ambient conditions but also in saline medium. A trained and optimized multinomial classifier (Error Correcting Output Codes) showed an acceptable precision and recall for an independent validation dataset. Even considering the strong absorbance of saline, the calculated geometric mean was 94 %, 87 %, and 71 % for oxalates, urates, and phosphates, respectively. A first real-time approbation during a real surgery (percutaneous nephrolithotomy) demonstrated a compatibility of the suggested approach with the surgical protocols and a good agreement of the acquired NIR spectra and the results of reference X-ray phase analysis.