Classifying SPECT Bone Metastasis Images in Grayscale Format with VGGNets

摘要

For the purpose of accurate diagnosis of bone metastasis diseases with nuclear medicine SPECT imaging technology in this work, we propose to study and construct CNN based classification model for identifying bone metastases with real-world SPECT imaging data. First, the geometric transformation including image mirroring, rotation, and translation is used to augment dataset of SPECT bone scans. Second, the standard VGG model is adopted to develop classifier of SPECT images. Lastly, a set of real-world data of whole-body SPECT bone scans is used to evaluate the built model, obtaining a best value of 0.996, 0.995, 0.997 and 0.995 for accuracy Acc, precision Pre, recall Rec and F-1 score are, respectively.