Hyperspectral imaging for tumor detection in the resection surface of fresh lumpectomy specimens (Conference Presentation)

Abstract

Real-time intra-operative resection margin assessment during breast-conserving surgery (BCS) is important to prevent incomplete tumor removal. Nowadays, in up to 37% of the women undergoing BCS, tumor positive margin are found after surgery. We test the feasibility of hyperspectral imaging to predict these positive margins in fresh lumpectomy specimens, to prevent incomplete tumor removal. Hyperspectral diffuse reflectance images (900-1700 nm) were collected on fresh lumpectomy specimens. These specimens were obtained from women undergoing primary BCS, that did not get neo-adjuvant treatment. To ensure hyperspectral images of the entire resection surface, we treated the specimen as a cube and imaged it from six different sides. Next, a SVM classification algorithm, which we developed and tested with a different dataset, was applied to these hyperspectral images to predict positive margins. Finally, we compared the margin assessment performed with hyperspectral imaging with histopathology, the gold standard for margin assessment. It was found that hyperspectral imaging could be used in the clinical workflow. First, data acquisition of the entire resection side was fast and took only 20 seconds per resection side. Second, with the earlier developed classification algorithm, data analysis could be performed in the operating theater in limited amount of time. Third, with hyperspectral imaging we were able to find 12 out of 13 positive resection sides. The one positive resection side that was missed contained a single malignant pocket smaller than 1 mm2. These preliminary findings make hyperspectral imaging a promising technique for resection margin assessment during BCS.