This paper presents the preliminary results of a deep learning model that aims to assist in the diagnosis of coronary artery disease in computed tomography images. In this approach, we use a convolutional neural network (CNN) classifier for the detection of abnormal images, that indicate the presence of coronary artery disease. A public dataset with 205 labeled cases of cardiac CT images were used. For the methods, we created an environment for the application of the CNN, prepared data preventing leaks, and dealt with class imbalance; At first, we used pre-trained models and fine-tuned the model using the CT images dataset. We divided our dataset into training, testing, and validation. Five different architectures were tested: MobileNetV2; Densenet201; Resnet, Nasnet, and EfficientnetV2. The classification performance of each model was evaluated through a 5 fold cross-validation by estimating the Area Under the Curve for the Receiver Operating Characteristic curve, loss, and accuracy. Best results were achieved with the EfficientNetV2 model, with 0.545 of loss, 0.909 of accuracy, and 0.931 of AUC-ROC. We also used Grad-CAM analysis to demonstrate the most important region within each case that the model based its prediction. EfficientNetV2 proved to be robust, being accurate in almost all the normal cases. This CNN made a few wrong predictions in abnormal cases, its Grad-Cam analysis demonstrated that the arterial regions were always in the relevant region. The results demonstrate the capabilities of CNN models to correctly predict abnormal cases of coronary artery disease in computed tomography. The next step in this research will be to test our model on larger datasets.