In Vivo Confirmation of Prostate Tumor Burden During HDR Brachytherapy With a Combined Raman Spectroscopy and MRI Radiomics Approach.

PURPOSE/OBJECTIVE(S) TRUS-guided biopsies still yield a notable rate of sampling error for prostate cancer (PCa) and do not allow in situ characterization. Raman spectroscopy (RS) is a minimally invasive optical imaging modality allowing to measure tissue properties within seconds. In this study, we present the first in man results from a navigated RS optical probe used to confirm sites of tumor burden during HDR brachytherapy. MATERIALS/METHODS The study included 8 prospectively recruited subjects with PCa prior to HDR brachytherapy. An electromagnetic (EM) guidance system was used to navigate in vivo RS acquisition using MRI-TRUS image fusion, and integrated within the HDR brachytherapy procedure workflow. The optical device includes a dual wavelength laser source (680 and 785 nm), a spectrometer, a custom design probe, and an acquisition/processing software. For each case, deformable registration of mpMRI (with prostate, urethra and GTV segmentations) to interventional 3D TRUS reconstructed images (with prostate and urethra segmentations), enabled visualization of GTV with the navigation system. The EM tracked RS probe was inserted through a co-axial needle. Once in place, high wavenumber (HW) and fingerprint (FP) spectra were measured, a biopsy was taken at the corresponding site, and the location was recorded; then, the HDR brachytherapy procedure resumed. Biopsy samples were evaluated histologically to confirm presence of cancer. For the radiomics analysis, 8 first order and 8 GLCM image features were calculated on the MRI (T2W and ADC), extracted from the corresponding optically scanned sites, using open-source software. RESULTS In total, 26 sites were inspected with the RS probe and biopsied (range 2-5 per case). The pathology results identified 14 samples as cancer (Gleason score ≥ 7) and 12 samples as benign. On average the optical measurements took 45 seconds per site (range 23-90) and added 20 minutes (range 15-24) to the intervention. The spectra were pre-processed and labeled according to the pathology results. Neighborhood component analysis was applied to select the features that optimize the prediction performance, extracting 15 features from FP, 1 from HW and 11 from MRI radiomics. Classification models using a support vector machine were trained with a leave one out scheme for evaluation, using different feature combinations. Table 1 presents the average results over all folds. CONCLUSION We presented an image-guided system allowing the first in situ characterization of prostate lesions with RS during brachytherapy, in a manner integrated with the procedural workflow. The acquired spectra from benign and malignant regions show promising preliminary data, demonstrating the potential of the technique for real-time PCa confirmation. Finally, the combination of optical and MRI features yields an improvement in classification accuracy.