Abstract
Retroperitoneal soft tissue sarcoma (RSTS) is a rare type of cancer with limited treatment options. Achieving complete resection with negative margins is one of the most significant prognostic factors for RSTS survival. The UltraProbe is a handheld point probe Raman spectroscopy system that significantly decreases the imaging time compared to the probe systems currently used. This study aims to determine the performance of the UltraProbe in detecting STS in an in vivo environment during their resection. Thirty patients were recruited at Maisonneuve-Rosemont Hospital, Montreal, Canada. Raman spectra were acquired during STS resection using the instrument. A machine learning random forest classification algorithm was developed to predict the diagnosis associated with new Raman spectra: STS or healthy tissue. The classification of Raman spectra as well-differentiated liposarcomas or normal adipose tissue was performed with a sensitivity of 94%, specificity of 95%, and accuracy of 94%. The classification of spectra as well-differentiated and dedifferentiated liposarcomas or normal adipose tissue was performed with a sensitivity of 90%, specificity of 93%, and accuracy of 90%. The classification of spectra as non-liposarcoma STS or protein-rich non-adipose tissue was performed with a sensitivity of 87%, specificity of 81%, and accuracy of 87%.