Recognition of oral squamous cell carcinoma is difficult due to delayed diagnosis and expensive data collection. An inexpensive and efficient computerized screening system is essential for an early diagnosis of the disease and to reduce the need for costly expert's analysis and intervention. Furthermore, transparency is crucial for these systems to align with important industrial applications. Explainable Artificial Intelligence (XAI) provides techniques for understanding models. In particular, an explanation focused on clinical users is fundamental. Among numerous XAI techniques, we propose a solution consisting in an Informed Deep Learning (IDL) approach to embed medical information in the system and the Case-Based Reasoning paradigm to produce visual explanations in the output. State-of-the-art architectures are used for comparative analysis of the classification model and a triplet-net is developed for modeling the feature space according to the medical knowledge. We collaborated with medical facilities to obtain a dataset for which we performed a series of experimental benchmarks. We measure the performance both in terms of classification accuracy and human-centered explainability.