The advent of data-driven science and artificial intelligence (AI) has provided a deeper knowledge about data that has driven the clinical research to unprecedented change. AI has shown great potential in the scoliosis diagnosis for which the current widely adopted standard of evaluation is the manual measurement on the X-ray radiographs of the Cobb angle to quantify the magnitude of spinal deformities in scoliosis. The reliability of the Cobb angle measurement mainly depends on the subjective experience of the operators and it is time-consuming. Machine learning (ML) and Deep Learning (DL) methods can help surgeons to avoid misjudgment about scoliosis screening, diagnosis and classification by providing a powerful solution for saving time and effort in the Cobb angle measurement. The contribution of this work is twofold. Primarily it aims to provide an overview of the main ML approaches, with special focus on DL, that can be used in the spine field to make physicians and researchers aware of the benefits of the ML approaches for scoliosis diagnosis and treatment with respect to the traditional methods. Furthermore, because the reliability of the all ML approaches depends strongly on the training data and often it is difficult to obtain a large amount of representative data, we survey the main databases containing spinal data with the purposes of providing the stakeholders with a knowledge that can substantially help them to choose the best dataset in the spine field for training their ML models. Furthermore, with this work we intend to lay the foundations for development of a ML-based tool for supporting the physicians in the diagnosis, classification, screening and prognosis prediction of scoliosis.