High-Quality Interpolation of Breast DCE-MRI Using Learned Transformations

Dynamic Contrast Enhancement Magnetic Resonance Imaging (DCE-MRI) is gaining popularity for computer aided diagnosis (CAD) of breast cancer. However, the performance of these CAD systems is severely affected when the number of DCE-MRI series is inadequate, inconsistent or limited. This work presents a High-Quality DCE-MRI Interpolation method based on Deep Neural Network (HQI-DNN) using an end-to-end trainable Convolutional Neural Network (CNN). It gives a good solution to the problem of inconsistent and inadequate quantity of DCE-MRI series for breast cancer analysis. Starting from a nested CNN for feature learning, the dynamic contrast enhanced features of breast lesions are learned by bidirectional contrast transformations between DCE-MRI series. Each transformation contains the spatial deformation field and the intensity change, enabling a variable-length multiple series interpolation of DCE-MRI. We justified the proposed method through extensive experiments on our dataset. It produced a more efficient result of breast DCE-MRI interpolation than other methods.