Feed

Hyperspectral imaging is a non-invasive imaging modality showing potential in delineating tumour margins preoperatively. This pilot study evaluated the feasibility of using hyperspectral imaging to demarcate lateral margins of high-risk facial basal cell carcinomas (BCC) prior to Mohs micrographic surgery. Thirty patients with high-risk BCCs were recruited from the Department of Dermatology, Sahlgrenska University Hospital, Sweden. Lesions were initially demarcated using dermoscopy, followed by hyperspectral imaging scans. During the first stage, a superficial vertical incision was performed along the demarcation line before adding a 3-mm clinical margin for the bowl-shaped excision of the tumour. Hyperspectral imaging-based tumour margins were compared with histopathologically verified borders, serving as ground truth. The data analysis used supervised learning; 2 complementary validation strategies were employed: a half-split approach where the left half of each annotated image was used for training and the right half for testing, and a leave-one-out cross-validation at the image level. A pixel-wise classification approach was used, treating each pixel as an independent sample. Hyperspectral imaging achieved a pixel-wise classification accuracy of 0.76, sensitivity of 0.75, specificity of 0.78, and an area under the receiver operating characteristic curve of 0.84. Hyperspectral imaging demonstrated potential for tumour demarcation, providing a basis for future research.