OSCC Diagnosis by Deep Learning-Case-Based Reasoning: a Pilot Study

Objectives: Oral Squamous Cell Carcinoma (OSCC) represents a significant global health concern with severe implications for patients. Despite oral cavity accessibility, many general dentists struggle to identify early OSCC lesions, leading to delay diagnosis and less favorable prognosis.
This matter has encouraged researchers to seek tools for aiding less experienced professionals in order to promptly identify OSCC lesions. Recently, by means of AI with Deep Learning (DL), clinical image analysis seems to be enhanced levels of recognition and classification of lesions.
Moreover, the integration of Case-Based Reasoning (CBR) could represent a promising approach for early detection of OSCC. It is an AI paradigm addresses new challenges by retrieving information from previously solved similar cases and adapting solutions to new contexts.
Our goal is to evaluate efficiency of a DL-CBR model for detection and classification of neoplastic, traumatic, and aphthous ulcers.
Methods: A total of 70 clinical images of oral lesions were collected at the Oral Medicine Unit of the University Hospital P. Giaccone in Palermo, Italy. For the training phase of DL-CBR model, we used a subset of 45 images, including 15 neoplastic ulcers, 15 aphthous ulcers, and 15 traumatic ulcers. Subsequently, the DL-CBR model was tested with 25 images (7 neoplastic, 12 aphthous, and 6 traumatic ulcers), thanks to the support of two oral medicine experts (blended phase).
Results: The DL-CBR model seems to achieve a high level of accuracy in classification and detection, identifying correctly 7/7 neoplastic, 11/12 aphthous and 6/6 traumatic ulcers.
So, remarkable results are achieved with a sensitivity of 96% and a specificity of 93.5% for all lesions.
Conclusions: The DL-CBR model seems to offer a powerful and reliable solution for detection and classification, at least for ulcerative lesion, with the added benefit of improving user understanding and confidence in diagnostic process.