Imaging breast cancer morphology using probe-based confocal laser endomicroscopy: towards a real-time intraoperative imaging tool for cavity scanning

Abstract

Current techniques for assessing the adequacy of tumour excision during breast conserving surgery do not provide real-time direct cytopathological assessment of the internal cavity walls within the breast. This study investigates the ability of probe-based confocal laser endomicroscopy (pCLE), an emerging imaging tool, to image the morphology of neoplastic and non-neoplastic breast tissues, and determines the ability of histopathologists and surgeons to differentiate these images. Freshly excised tumour samples and adjacent non-diseased sections from 50 consenting patients were stained with 0.01 % acriflavine hydrochloride and imaged using pCLE. All discernible pCLE features were cross-examined with conventional histopathology. Following pattern recognition training, 17 histopathologists and surgeons with no pCLE experience interpreted 50 pCLE images independently whilst blinded to histopathology results. Three-hundred and fifty pCLE image mosaics were analysed. Consistent with histopathology findings, the glandular structures, adipocytes and collagen fibres of normal breast were readily visible on pCLE images. These were distinguishable from the morphological architecture exhibited by invasive and non-invasive carcinoma. The mean accuracy of pCLE image interpretation for histopathologists and surgeons was 94 and 92 %, respectively. Overall, inter-observer agreement for histopathologists was 'almost perfect', κ = 0.82; and 'substantial' for surgeons, κ = 0.74. pCLE morphological features of neoplastic and non-neoplastic breast tissues are readily visualized and distinguishable with high accuracy by both histopathologists and surgeons. Further research is required to investigate a potential role for the use of pCLE intraoperatively for in situ detection of residual cancerous foci, thereby guiding operating decision-making based on real-time breast cavity scanning.

Unsupervised Feature Learning for Endomicroscopy Image Retrieval

“…The dataset is collected by a pre-clinical pCLE system (Cellvizio, Mauna Kea Technologies, Paris, France) as described in [11]. Breast tissue samples are obtained from 45 patients that are diagnosed with three main classes including normal, benign and and neoplastic.…”

Methylene-blue aided rapid confocal laser endomicroscopy of breast cancer

“…A preliminary study on 71 ex-vivo breast tissue samples imaged with the Cellvizio pCLE system (Mauna Kea Technologies) demonstrated that cancerous and normal tissue could be differentiated by histopathologists with an accuracy of 94%. 10 This study used topically applied acriflavine, excited at 488 nm, which stains cell nuclei and enables high contrast imaging of tissue morphology. However, this is not currently an attractive option for in-vivo imaging due to the lack of regulatory approval and suggested risks of mutagenicity.…”

“…The high frame rate enables image mosaics to be assembled more readily, 21 which, as discussed in Ref. 10, is important for allowing features to be identified with confidence.…”

“…The accuracy of normal/tumor identification using acriflavine staining pCLE has already been established; 10 so, the main purpose of this pilot study was to determine whether the 660-nm system could identify features following the taxonomy of pathologies developed in the previous study. A total of 450 pCLE image mosaics (mean 30 per sample; range 20 to 40) were generated from 15 specimens.…”

“…The potential of the approach to identify the microscopic tissue morphology relevant to tumor identification is demonstrated by comparison of the reconstructed mosaics with those from the previously generated data bank generated using the Cellvizio pCLE with acriflavine (0.02%) staining, and with conventional histology. 10 Two features typifying normal breast tissue were identified: adipose tissue and the presence of collagen fibres in the stroma. As shown in Figs.…”