MSACL 2023 Abstract

Introduction
Rapid evaporative ionization mass spectrometry (REIMS) can profile and classify tissues without sample preparation in real-time, using pre-built spectral databases and machine-learning. REIMS has been proposed as an intraoperative tool to inform surgeons during breast tumor removal, to reduce the need for re-operation due to remaining cancer cells on the periphery (positive margins), which occurs in approximately 25% of cases. While REIMS in its current form can indicate the presence of cancerous tissue, it does not indicate 3D location. This limits the ability to validate mass spectra obtained intraoperatively, and in the future, its ability to inform decision-making in the operating theatre.

Objectives
In the current study, we combined REIMS with a spatio-temporal navigated cautery that was temporally-synched to REIMS. We tested this platform intraoperatively on 22 breast cancer (BCa) surgery cases, and retrospectively compared the margin status determined by the navigated REIMS system with that of the ‘gold standard’ histopathology report.

Methods
A multivariate model based on PCA/LDA was trained and cross-validated on REIMS spectra sampled from 11 pathology-validated ex vivo BCa surgery specimens, including N=118 spectra from normal breast adipose, and N=36 spectra from invasive BCa. During 22 surgeries, REIMS was used to acquire testing data from tissue dissected with an electromagnetically-tracked cautery, and surgeons would call-out tissue types being dissected. Spectra were retrospectively classified using the above model. Prior to surgery, a localization wire with an electromagnetic sensor was placed into the tumor as a marker and a 3D map of the tumor was contoured using tracked ultrasound. REIMS was temporally-synched with the 3D tracker, allowing computation of the spatial origin of spectra using the time of acquisition. Spectra classified as BCa were mapped onto a display of the tumor region and compared with the pathology report.

Results
Spectra from ex vivo BCa specimens were characterized by a distinctly elevated ratio of glycerophospholipids (PL)-to-triglycerides (TG) as compared with normal breast adipose. A PCA/LDA-based classifier exhibited >90% accuracy on cross-validation. In the intraoperative patient cohort, 4/22 cases had positive margins as assessed by pathology, all of which were correctly identified by REIMS. Notably, two of these cases were positive for ductal carcinoma in situ with the location of the positive margin matching the pathology report. Intra-operative spectra classified as BCa exhibited similar PL:TG ratios to that observed in the ex vivo data. Further, 18/22 cases were margin negative, and observations from 12 of those cases were consistent with the histopathology assessment. There were 6 margin-negative cases that were determined to have positive margins by REIMS (‘false positive’). As well, 4/6 false positive cases were noted by histopathology as ‘close margins’ with cancer cells being detected within 1 mm of the inked margin or noted by the surgeon as containing dense but otherwise normal breast tissue. Other normal tissue types dissected during breast surgery with high PL content were misclassified as BCa, including skin and muscle, but these spectra can be rationalized by relative distance from the tumor region using the navigation data and/or call-outs from the surgeon.

Conclusions
Our study points to the importance of spatio-temporal tracking to help validate intraoperative tissue characterization by REIMS profiling during BCa surgery. We anticipate that a navigated REIMS platform will eventually help inform the surgeon of specific locations where a wider excision may be necessary to avoid a positive margin.