Abstract Cancers of the upper gastrointestinal tract remain a major contributor to global cancer risk. Accurate resection margin assessment is crucial for improved survival and reduction in recurrence. Diffuse reflectance spectroscopy (DRS) is a point-based optical technique that allows discrimination of tissue type. The aim of this study was to assess the diagnostic accuracy of a DRS probe integrated with a real-time tracking system to differentiate tissue types in vivo to aid margin assessment. Patients undergoing elective oesophageal and gastric cancer resection surgery at a tertiary hospital in London were prospectively recruited. A hand-held sterilisable DRS probe was used to acquire spectra from normal and tumour tissue intra-operatively. Binary classification was achieved using supervised machine learning classifiers, which were evaluated in terms of sensitivity, specificity, accuracy and the area under the curve. A total of 9586 spectra were collected from 32 patients. For stomach tissue, the XGB classifier achieved sensitivity and specificity of 77 and 95%, respectively, with a diagnostic accuracy of 90.5% when differentiating normal and tumour tissue. For oesophageal tissue, the LGBM classifier highlighted sensitivity and specificity of 86 and 82%, respectively, with an accuracy of 84%. The AUC exceeded 92% in both datasets. Our results demonstrate that DRS, combined with real-time tracking and machine learning, can differentiate normal and tumor oesophageal and gastric tissue with high diagnostic accuracy in vivo. This study presents a novel, clinically applicable method that integrates engineering innovation with surgical practice with the potential of improving intraoperative decision making and margin assessment.
Nazarian et al. (Sun,) studied this question.