Detection of Periapical Lesions on Dental Panoramic Radiographs Based on Artificial Intelligence

Dental periapical damages can have various reasons and is reflected by a radiolucent lesion on complementary imaging: angulated retro-alveolar (RA) radiographs, dental panoramic radiographs, and three-dimensional imaging such as computed tomography (CT) or cone-beam computed tomography (CBCT). For the radiographic detection of these deep periodontal lesions, the dental panoramic represents a first approach commonly performed with relatively low radiation. The investigation can be followed by retroalveolar radiology imaging that are more localized and more precise. However, using these techniques, the detection rates of these lesions are low (20% and 36% respectively), it is necessary to use three-dimensional tomographic investigation to be more discriminating (69%). The gold standard imaging for detection of these lesions is CBCT followed by retroalveolar radiography (\~2x less sensitive than CBCT) and panoramic radiography (\~2x less sensitive than RA). Although not a full-thickness radiograph, the dental panoramic has the advantage of being more commonly performed while being less radiating than CBCT and giving a global view of the dental arches on a single image. The detection of periapical lesions is done after a clinical assessment and a visual appreciation of the complementary examinations. The aim of this project is to improve the detection of periapical lesions, by developing an algorithm able to identify them on a panoramic dental radiograph. This algorithm is based on a deep learning system trained with reference data including panoramic dental imaging and CBCT with an acquisition interval of less than 3 months. The model is based on a previous work, will improve the quality of the initial data (using CBCT), using innovative artificial intelligence algorithms (transfer learning).