Raman optical identification of renal cell carcinoma via machine learning.

High pathologic tumor-node-metastasis (pTNM) stage grade or Fuhrman grade indicates poor oncological outcome in renal cell carcinoma (RCC). Early diagnosis and screening of these RCCs and adjust surgical planning accordingly are greatly beneficial to patients. Raman spectroscopy is a highly specific fingerprint spectrum on molecular level, pretty appropriate for label-free and noninvasive cancer diagnosis. In this work we established a Raman spectrum-based supporting vector machine (SVM) model to accurately ex vivo distinguish human renal tumor from normal tissues and fat with an accuracy of 92.89%. The model can also be used to determine tumor boundary, showing consistent results to pathological staining analysis. This method can be additionally used to accomplish the classification purposes of renal tumor subtypes and grades with an accuracy of 86.79% and 89.53%, respectively. Therefore, we prove that Raman spectroscopy has great potential in the rapid and accurate clinical diagnosis of renal cancers.