Nowadays it is possible to use neural networks for medical applications, ranging from the prediction and diagnosis of diseases to the determination of the most appropriate treatment for patients. The aim of this article is to present a systematic review of scientific articles related to neural networks used for disease prediction, with the purpose of analyzing, by means of a systematic review, the models used and to demonstrate their prediction accuracy. PRISMA methodology was used to analyze 121 relevant articles, inclusion and exclusion criteria were applied and finally 38 remained. As results, 38 studies (2016–2024) using neural networks for prediction of 26 diseases are presented; CNNs prevail and, together with ANN, LSTM and hybrid models, achieve accuracies >95% (some cases ≈ 100%). 80% of the publications date from 2022 to 2024, 84% come from Asia and are disseminated in 28 journals. The SWOT points out as strengths: high throughput and early detection; weaknesses: black box, need for big data and ethical regulatory challenges, so it is concluded that the incorporation of neural networks brings significant benefits both in the prediction and treatment of diseases.