IADR Abstract Archives
A Novel Multi-layer Artificial Intelligence Neural Network for Diagnosis of Orofacial Pain.
Objectives: Orofacial pain and temporomandibular disorders (TMD) constitute one of the most prevalent diseases in the general population. However, patient access to appropriate diagnosis and care is difficult, usually experiencing multiple missed diagnosis and unnecessary treatments. Artificial intelligence was shown to help clinicians in the diagnosis of other diseases. The present study developed a novel artificial intelligence system for the diagnosis of orofacial pains and TMD and compared its accuracy to dental clinicians.
Methods: We developed an artificial neural network, multi-layer perceptron (MLP), consisted of a machine learning model that simulated in many ways the function of the brain. It was constructed based on individual artificial neurons that included an input layer, 5 hidden layers and one output layer (Fig). It was trained using a backpropagation algorithm.
The study also compared the capability of the MLP vs. dental professionals to diagnose different categories of clinical situations, including acute pain from dental origin, orofacial pain from cardiac origin, temporomandibular joint dysfunction, trigeminal neuralgia and others. The data of each clinical case was entered to the MLP and also presented to dental clinicians (n=8) and a diagnosis was obtained. A generalized mixed model framework was used for data analysis. Statistical significance was determined using a 0.05 level. Ethical approval was obtained.
Results: Overall, the diagnostic accuracy of the MLP was superior to dental clinicians (p=0.0053) and this was more evident in those clinical cases that involved non-odontogenic pain (e.g. neuropathic pain, cardiac pain, neurovascular pain).For example, only two out of eight clinicians were able to diagnose those cases of orofacial pain that was the only symptom of a cardiac ischemia event while the MLP was accurate. Furthermore, none of the clinicians were able to diagnose those cases of orofacial pain from neurovascular origin while the MLP did.
Conclusions: This is the first study to develop and test artificial intelligence that can diagnose orofacial pains and TMD. It can be used to assist medical and dental clinicians. In some cases, the MLP appears to have a life-saving role. Further research will test the network potential in real clinical settings.
Methods: We developed an artificial neural network, multi-layer perceptron (MLP), consisted of a machine learning model that simulated in many ways the function of the brain. It was constructed based on individual artificial neurons that included an input layer, 5 hidden layers and one output layer (Fig). It was trained using a backpropagation algorithm.
The study also compared the capability of the MLP vs. dental professionals to diagnose different categories of clinical situations, including acute pain from dental origin, orofacial pain from cardiac origin, temporomandibular joint dysfunction, trigeminal neuralgia and others. The data of each clinical case was entered to the MLP and also presented to dental clinicians (n=8) and a diagnosis was obtained. A generalized mixed model framework was used for data analysis. Statistical significance was determined using a 0.05 level. Ethical approval was obtained.
Results: Overall, the diagnostic accuracy of the MLP was superior to dental clinicians (p=0.0053) and this was more evident in those clinical cases that involved non-odontogenic pain (e.g. neuropathic pain, cardiac pain, neurovascular pain).For example, only two out of eight clinicians were able to diagnose those cases of orofacial pain that was the only symptom of a cardiac ischemia event while the MLP was accurate. Furthermore, none of the clinicians were able to diagnose those cases of orofacial pain from neurovascular origin while the MLP did.
Conclusions: This is the first study to develop and test artificial intelligence that can diagnose orofacial pains and TMD. It can be used to assist medical and dental clinicians. In some cases, the MLP appears to have a life-saving role. Further research will test the network potential in real clinical settings.
3640333_File000000.jpg