CNN-Based Classifier For Diagnosing SPECT Bone Images

To assist nuclear medicine doctors for quickly and accurately identifying bone metastases, in this work, we develop a convolutional neural network to automatically classify whole-body SPECT bone scans. First, the original data of SPECT bone scans are augmented by using geometric transformation. Second, two deep model VGG5 and VGG7 are developed by following the famous VGG model, where a 2×1 maximum pooling is introduced before the first layer convolution. Experimental evaluation conducted on real-world SPECT images shows that the developed models are workable to classify SPECT images into categories of concern.