IntroductionThe mechanism of tumor generation and its function remains unknown. Also, metastasis, which is the spread of tumors in the body, is one of the researchers' concerns. The diagnosis of these abnormalities with different shapes and sizes is critical to the decision-making process; Especially when the metastasis occurs in the brain. Therefore, designing and utilizing an automated detection method and tracking can be helpful for clinicians. One of these methods named the Convolutional Neural Network (CovNet/CNN), is a deep learning algorithm that can take in an input image, assign importance (learning rates and biases) to various objects/aspects differentiate one from the other. In general, these kinds of networks are mainly composed of three different sub-networks including convolutional layers which have various types of filters that each filter recognizes a specific image texture/shape pattern. Pooling layers, which are responsible for dimension reduction to reach less computational complexity, and fully-connected layers which di the classification task. Therefor, in this study, we propose a deep learning method for tracking using a convolutional neural network (CNN).MethodThis study used three-dimensional T۱-weighted MPRAGE MRI data of ۷۴ subjects with brain metastasis from breast, lung, prostate, and melanoma. We scanned these subjects in the Imaging Center of Imam Hossein Hospital. Aiming to track metastasis as an object in each slice, we used a ten folded CNN classifier with ۵۶ data for the train and ۱۸ data for tests after quality assurance and preprocessing. We compare our design with two popular methods Siam-FC and RT-MDNets.ResultsAfter the implementations, the algorithm made a prediction and tracing model for the brain metastasis. The three well-known evaluation quantities in tracking concepts are "Success", "Precision", and "Frame Per Second" (FPS). The results are presented in Table ۱. Besides, frame by frame detection for a subject with one metastasis is shown in Figure ۱.ConclusionsWe designed a model based on CNN that could track metastasis. This model can help radiologists diagnose tumors and reduce human error as much as possible. Although this algorithm still has shortcomings in detecting the initial slice, it works well in tracing and following the subject. Further studies with large sample sizes are warranted to an improved simulation of the system under study by the trained CNN.