Light-reflectance spectroscopy to detect positive surgical margins at radical prostatectomy: exploration of new algorithms to refine detection rate
Sources of Funding: none
Introduction
Light-reflectance spectroscopy (LRS) is a novel technology which can reliably detect positive surgical margins (PSMs) for Gleason score ≥7 prostate cancer in ex vivo radical prostatectomy (RP) specimens. Furthermore, LRS can provide near-immediate feedback to the surgeon and potentially influence surgical decision making. Our objective was to refine our technique using a new algorithm to increase the efficacy and performance of this technology.
Methods
A prospective evaluation of ex vivo RP specimens using LRS was performed at a single institution from March 2016 to October 2016. LRS measurements were performed on selected sites on the prostate capsule, marked with ink, and correlated with pathological analysis. The previous 5 feature algorithm, which has been validated using both a training and testing set, was further optimized with an additional 4 features. This new 9 feature algorithm included additional shifts in slope at specific points along the 690 nm - 770 nm wavelength range to better differentiate malignant from benign tissue. The sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV), and area under the receiver operating characteristic curve (AUC) for LRS predicting PSMs using the 5 and 9 point algorithms were calculated._x000D_
Results
197 sites in 34 RP specimens were evaluated by LRS and histopathology. The 5 and 9 point algorithms behaved similarly for Gleason groups 1-2 showing poor sensitivity (33%,33%), PPV (18%,27%), and AUC (0.54,0.59). Specificity was better in the 9 compared to the 5 point algorithm (85.3 vs. 74.4%, respectively). Both algorithms performed much better for Gleason groups 3-4. Specificity was improved for the 9 point algorithm compared to the 5 (90.6 vs. 84.4%, respectively). Sensitivity was improved for the 5 point algorithm compared to the 9, (100 vs. 77.8%, respectively). Both algorithms had great NPV for low grade and high grade cancer.
Conclusions
Incorporating features of both LRS algorithms improves performance to reliably detect PSMs of higher grade prostate cancer. This is likely secondary to increased cellular density in these cancers causing more differential light scattering. Further refinements and additional prospective evaluations may allow incorporation into clinical practice.
Funding
none
Noah Canvasser
Xinlong Wang
Henry Chan
Hanli Liu
Payal Kapur
Claus Roehrborn
Jeffrey Cadeddu