<p><strong>Background:</strong> Thyroid Nodules are more and more widespread, with the frequency of thyroid cancers increasing accordingly. In routine practice, the practician's goal is to determine if a nodule is malignant or suspicious, and thus requires surgery. However, nearly one-third of thyroid nodules fall into the 'indeterminate' classification segment, often leading to the use of surgery as a diagnostic tool, when it could have been avoided with a better preoperative diagnosis. Molecular tests exist to address this issue, but they are markedly onerous.</p>
<p><strong>Results:</strong> In this paper, we present an accurate, cost-less method to predict the malignancy of thyroid nodules using a categorical thyroid nodules dataset. The prediction is achieved through the use of a new supervised autoencoder neural network. Results show that the Supervised Autoencoder Neural Network outperforms classical classification methods on all evaluation metrics. A proof-of-concept Graphical User Interface was developed to demonstrate the medical usefulness of the latent space of the trained network.</p>
<p><strong>Conclusions:</strong> Aside from the prediction itself, the Supervised Autoencoder Neural Network provides three medically interesting characteristics: a prediction confidence score, consistent feature ranking, and an interpretable latent space. All of those characteristics are valuable from the practician's point of view, as they allow for an efficient, cost-less and well-informed prognosis.</p>