Introduction

The number of small, incidentally detected renal masses increase steadily. About 6,000 benign cases are misclassified radiographically as malignant and removed surgically yearly. Raman spectroscopy (RS) has been widely demonstrated for tissue characterization, however current implementations with either 785 or 830 nm near-infrared excitation have been ineffectual in tissues with intense auto-fluorescence such as the kidney. Recently, a RS system using 1064 nm light source was described which may have greater sensitivity for malignant and benign tissue discrimination due to decreased bulk tissue auto-fluorescence. Our aim was to evaluate the ability of 1064nm RS to distinguish normal and malignant renal tissue.

Methods

Ex vivo specimens of Renal Cell Carcinoma and healthy human kidney were obtained from the Cooperative Human Tissue Network at Vanderbilt university. Measurements were made using of a benchtop dispersive 1064 nm Raman system. Multiple spectra were acquired from at least 5 physical locations across each specimen. A total of 93 measurements were used for the final analysis. _x000D_ The resulting spectra were put into a machine learning algorithm, sparse multinomial logistic regression (SMLR), to predict class membership of healthy and malignant tissues, and cross-validated using a leave-one-specimen out approach. Posterior probabilities of group classifications were extracted. Spectral bands that robustly differentiated between malignant and benign tissue were identified by the SMLR algorithm. A quantitative metric based on SMLR outputs called feature importance, defined as the product of the mean weight and frequency of usage of each feature, guided the association of spectral features with biological indicators of healthy and diseased Kidney tissue. _x000D_

Results

The SMLR algorithm identified 152 significant Raman spectral bands. most important features are depicted in figure 1. Correct classification by the SMLR algorithm was obtained in 93.33% of the trials with sensitivity, specificity, negative and positive predictive value of 93.2%, 88.6, 92.9% and 89.2% respectively.

Conclusions

RS can accurately differentiate normal and malignant renal tissue. This suggests implications for utilizing RS for optical biopsy and surgical guidance in nephron sparing surgery.

Funding

none