Micro-morphological feature visualization, auto-classification, and evolution quantitative analysis of tumors by using SR-PCT

Tissue micro-morphological abnormalities and interrelated quantitative data can provide immediate evidences for tumorigenesis and metastasis in microenvironment. However, the multiscale three-dimensional nondestructive pathological visualization, measurement, and quantitative analysis are still a challenging for the medical imaging and diagnosis. In this work, we employed the synchrotron-based X-ray phase-contrast tomography (SR-PCT) combined with phase-and-attenuation duality phase retrieval to reconstruct and extract the volumetric inner-structural characteristics of tumors in digesting system, helpful for tumor typing and statistic calculation of different tumor specimens. On the basis of the feature set including eight types of tumor micro-lesions presented by our SR-PCT reconstruction with high density resolution, the AlexNet-based deep convolutional neural network model was trained and obtained the 94.21% of average accuracy of auto-classification for the eight types of tumors in digesting system. The micro-pathomophological relationship of liver tumor angiogenesis and progression were revealed by quantitatively analyzing the microscopic changes of texture and grayscale features screened by a machine learning method of area under curve and principal component analysis. The results showed the specific path and clinical manifestations of tumor evolution and indicated that these progressions of tumor lesions rely on its inflammation microenvironment. Hence, this high phase-contrast 3D pathological characteristics and automatic analysis methods exhibited excellent recognizable and classifiable for micro tumor lesions.