Live WebinarPeriodontal and peri-implant diseases: non-surgical protocols
14 Jul 2020, 08:00 PM Berlin
The study was a collaboration between researchers at the Finnish Center for Artificial Intelligence, Tampere University Hospital in Finland, Finnish manufacturer Planmeca and the Alan Turing Institute in the UK. In the study, the researchers developed a novel deep learning method that helps automatically determine the exact location of mandibular canals. The model is based on training and using deep neural networks, employing a dataset consisting of CBCT scans.
After training the model on the coarsely annotated volumes, the researchers were able to accurately localise the mandibular canals of the voxel-level annotated set, the mean curve distance and average symmetric surface distance being 0.56 mm and 0.45 mm, respectively. The results show that the model successfully outperformed the statistical shape models typically used in research.
According to the researchers, the new model can achieve near-human accuracy in cases in which the patient does not have any pre-existing conditions and does not require special treatment. “In more complex cases, one may need to adjust the estimate, so we are not yet talking about a fully stand-alone system,” said lead author Joel Jaskari, a doctoral candidate at Aalto University in Finland, in a press release.
The researchers noted that the aim of the study was to optimise the workflows of radiologists. “The aim of this research work is not, however, to replace radiologists but to make their job faster and more efficient so that they will have time to focus on the most complex cases,” explained Prof. Kimmo Kaski, senior adviser in computational science at Aalto University.
Planmeca, which specialises in developing 3D and 2D digital imaging devices, dental units, and CAD/CAM solutions and software, is currently integrating the model into its dedicated software. The model will be used with Planmeca 3D tomography equipment.
The study, titled “Deep learning method for mandibular canal segmentation in dental cone beam computed tomography volumes”, was published online on 3 April 2020 in Scientific Reports.