Mieke Bus

5 71 OCT analysis OCT datasets were subjected to in-house developed staging and grading protocols by two experienced OCT investigators (MTJB, DMdB) blinded for pathology results, pre-operative imaging or other clinical information assessed the OCT images. Each OCT image was visually inspected and classified as normal (without visible lesion) or abnormal (with visible lesion). For staging, visible lesions were divided in invasive and non-invasive as described before. (4) In short, full 3D OCT datasets were visually analysed on the appearance of a layered struc- ture underneath a lesion, loss of anatomical architecture, intensity of the first layer appear- ance and structural differences in the A-line. Lesions suspicious for CIS were classified as non-invasive. For grading, the decrease of light intensity [mm -1 ] was quantified after imperatively account- ing for system specific calibrations. (9, 10) The decrease of light intensity was determined by fitting OCT data in a manually selected region of interest in a lesion, to a single exponen- tial decay model in order to determine µ oct as described before. (11-13) From each patient we selected five 2D cross-sectional OCT scans from the middle part of clearly visible lesions. Statistics Statistical analysis was performed using MedCalc® Version 14.12.0 (MedCalc Software bvba). The statistical setup of the analysis was performed in collaboration with the Clinical Research Unit of our Hospital. Grading Because of unequal sample size in all groups, Wilcoxon rank sum test was used to describe differences in µ oct between low grade (grade 1 and grade 2, low-grade) and high-grade (grade 2 high-grade and grade 3) lesions. Boxplots were used for visualization. Because of continuous data, ROC-analysis was performed to calculate sensitivity, specificity and the cut-off value for OCT grading. 2x2 table calculation was performed to determine sensitivity and specificity for tumor grading on biopsy. For analysis, non-diagnostic OCT scans and biopsies were excluded. For biopsy grading, non-diagnostic (benign/ND) was indicated when no tumor was present (e.g. due to limited sample size). For OCT grading, non-diagnostic was indicated when no attenuation coefficient could be extracted from an acquired dataset. Staging 2x2 table calculation was performed to determine sensitivity and specificity for tumour stage on OCT imaging and biopsy. Non-diagnostic OCT scans and biopsies were excluded.