Fractal analysis on the detection of the malignancy changes of pancreatic cancer

Abstract

Carcinoma of the pancreas is frequently associated with spatial changes between the centre and periphery of tumours. In order to convert histopathological objectives into new prognostic criteria for clinical needs, a semi-automated approach is investigated using fractal. Three different segments, known as the lumen, cell cytoplasem, and stromal tissue, of pancreatic malignancy were considered under this approach. Method concerning tissue preparation, morphological-segmentation, and applied parametric extraction were applied for examination and factor identification. For this purpose, samples from 21 cases of pancreatic carcinoma were stained, using the sirius red, light-green method. A number of 105 samples from each location (center/periphery) were obtained for examination as images in 512x512x3 pixel format. Segmentation was achieved using a clustering technique, and images were manually segmented into representative colors. For feature extraction, masked images were used and pre-processing approach of hue, saturation, and intensity color space was applied. Estimated fractal dimension were analyzed to examine whether the self-similarity features can be usefully implemented to detect changes in the tumour invasion between and within individual groups. Obtained results showed that fractal dimension is found on the malignancy of pancreases. This work supports scopes for the advantage of using automatic geometric analysis to define diagnostic markers on pancreatic tumour