Bone is one of the most invaded tissues by the metastases of breast, prostate, lung, kidney, and thyroid cancer. Patients with bone metastases are treated based on fracture risk: patients with a high fracture risk are considered for prophylactic surgery to prevent pathological fracture, whereas patients with low fracture risk are treated with conservative treatment such as radiotherapy for pain management.
Finite element (FE) models can be used to evaluate metastatic bone fracture risk. However, currently, the FE models are not yet applicable to femurs affected with osteoblastic metastases. To determine the mechanical properties within the FE model, calibrated bone densities from CT scans are obtained and since osteoblastic metastases appear very dense on CT scans this results in strong mechanical properties of the metastatic lesions although they are weaker in real life. A potential solution is to assign more appropriate material properties to the osteoblastic lesions to better simulate the mechanical behaviour of the weakened tissue. For this purpose, exact segmentation of the metastatic lesions is important.
Manual segmentation is a time-consuming task and it may be difficult to accurately segment every abnormality in the bone structure. The main goal of this study, therefore, is to automate the segmentation procedure using deep convolutional neural networks.
Researcher for this project: Ali Ataei.