Lesion Detection of Computed Tomography and Magnetic Resonance Imaging Image Based on Fully Convolutional Networks

Abstract

Computed tomography (CT) and Magnetic resonance imaging (MRI) are two kinds of important medical images, simply namely CT and MRI. Automatic lesion detection of CT and MRI is an important step for accurate clinical diagnosis. The classical CT and MRI lesion segmentation methods have bad performance due to the complex background noise, various illumination, and uneven color on CT image. In this paper, an improved fully convolutional network (FCN) model is proposed for lesion detection of CT and MRI image. The structure is same as FCN, and the lesion information from a deep layer is combined with appearance information from a shallow layer. First, we labeled all of the images from training set manually, the lesion and background labeled as 1 and 0, respectively. Then, the whole CT and MRI image dataset is fed to FCN. After 100 epochs training iterations, the model after the last iteration is selected as the final model, and then test dataset is put into the final model to obtain the detection results. The experimental results show that the proposed method can effectively detect and segment the lesion of CT and MRI images and greatly improve the segmentation accuracy, and can be used for the automatic lesion detection of CT and MRI images.