The staging diagnosis of multiple myeloma is conducive to the treatment of patients and can improve the survival rate of patients. In particular, the early diagnosis of multiple myeloma is more helpful to the cure of patients. However, multiple myeloma has a slow onset. Due to no obvious symptoms in the early stage, it is easily misdiagnosed. Recently, convolutional neural networks (CNNs) have been used in medical image detection to improve diagnosis efficiency. However, because of the slow onset of multiple myeloma in the early stage, how to segment diagnosis is a difficult point, which leads to less studies on the application of CNN algorithm in multiple myeloma. In order to fill this gap, a large number of existing multiple myeloma imaging data was used to construct a CNN model, and input the retained case image data into the constructed CNN to verify the accuracy of the neural network. The results showed that the accuracy rate of the neural network model constructed in this study was 0.87, which was higher than the accuracy rate of manual detection of 0.77. Therefore, it can be proved that the CNN model established in this paper is effective. At the same time, it is found that the use of magnetic resonance imaging (MRI) to classify and classify multiple myeloma has a high accuracy rate. Our results prove that the neural network algorithm can be applied to MRI analysis, which helps to improve the efficiency of multiple myeloma diagnosis.