Breast cancer is one of the most prevalent oncological diseases worldwide, and it has a significant impact on public health and the quality of life of those affected. This systematic review evaluates the efficiency of using Convolutional Neural Networks (CNN) for the diagnosis of breast cancer, comparing the studies in terms of sample size, performance factors, training time, and computational cost, among others. To this end, 232 studies were methodically identified in the Scopus and IEEE databases, of which 22 papers met the established inclusion and exclusion criteria. The results showed a predominance of the development of Classification models versus Segmentation models, as well as higher values of the performance factors in the models that used Magnetic Resonance Imaging (MRI) as input information. Therefore, it was concluded that MRI is more efficient in training CNN compared to mammography, since it reaches 100% Accuracy, Precision, F1-score, Sensitivity and Specificity, in addition, it only needs on average 17.6% of the sample size compared to mammography studies.