Deep Learning Based Lumbar Metastases Detection and Classification from Computer Tomography Images

Metastasis in the vertebral body is a widely observed malignant disease. This study used Computer Tomography (CT) images of the lumbar region to detect and classify the metastases from the normal vertebral bodies. We also classify the two types of metastases; Sclerotic and Lytic. Segmentation of the vertebral body is done using a deep neural network. We compared the performance of our model with existing state-of-art models. Multi-resolution blocks in the proposed model help segment the vertebral body with lytic lesions in the margin of the region of interest. Through Wavelet image transformation, coefficients are derived from the vertebral region. Based on the extracted wavelet scattering features, vertebral bodies are classified into three classes; normal, sclerotic and lytic, using Principal Component Analysis (PCA).