Ex-vivo freshly surgical removed pancreatic ductal adenocarcinoma (PDAC) specimens were assessed using pCLE and then processed for paraffin embeding and histopathological diagnostic in an endeavour to find putative image analysis algorithms that might recognise adenocarcinoma. Methods: Twelve patients diagnosed with PDAC on endoscopic ultrasound and FNA confirmation underwent surgery. Removed samples were sprayed with acriflavine as contrast agent, underwent pCLE with an experimental probe and compared with previous recordings of normal pancreatic tissue. Subsequently, all samples were subjected to cross-sectional histopathology, including surgical resection margins for controls. pCLE records, as well as corespondant cytokeratin-targeted immunohistochemistry images were processed using the same morphological classifiers in the Image ProPlus AMS image analysis software. Specific morphometric classifiers were automatically generated on all images: Area, Hole Area (HA), Perimeter, Roundness, Integrated Optical Density (IOD), Fractal Dimension (FD), Ferret max (Fmax), Ferret mean (Fmean), Heterogeneity and Clumpiness. Results: After histopathological confirmation of adenocarcinoma areas, we have found that the same morphological classifiers could clearly differentiate between tumor and non-tumor areas on both pathology and correspondand pCLE (area, roundness, IOD, ferret and heterogeneity (p < 0.001), perimeter and hole area (p < 0.05). Conclusions: This pilot study proves that classical morphometrical classifiers can clearly differentiate adenocarcimoma on pCLE data, and the implementation in a live image-analysis algorithm might help in improving the specificity of pCLE in vivo diagnostic.
Detecting Apples in the Wild: Potential for Harvest Quantity Estimation
