IADR Abstract Archives
Computer Assisted subclinical classification of TMJ OA
Objectives: To test novel quantitative tools based in 3D Shape Analysis of the Temporomandibular Joint (TMJ) to help characterize TMJ osteoarthritis (OA) phenotypes.
Methods: The study sample consisted of 91 healthy and TMJ OA patients. Our approach utilized Statistical Shape Modeling (SSM) to obtain information about TMJ OA phenotypes. All healthy subjects had radiographic diagnosis of no condylar pathology; TMJ OA patients underwent a clinical exam by an orofacial pain specialist to confirm the TMD research diagnostic criteria. All participants had a CBCT scan taken (i-CAT Next Generation, 120 kV, 18.66 mA). 3D models of the TMJs were generated from CBCT, and left joints were mirrored into the right side. Correspondent Point distribution models (PDM) were constructed using SPHARM-PDM. After quality control, the cohort comprised 218 TMJ joints (153 TMJ OA, 65 healthy). TMJ OA joints were by consensus subdivided into 7 subgroups based in pathological morphological phenotypes. We generated SSM-shape spaces to describe the morphology of each TMJ OA subgroup and the healthy group. Then we projected each TMJ OA case into each SSM-shape space, and shape loads were computed to calculate similarity with that space. A case will be classified in the group for which the similarity metric is smaller.
Results: 9.8% TMJ OA cases were misclassified. No TMJ OA case was classified in the healthy group. From the 15 out of 218 misclassifications, 12 had the correct classification as second closest. For the remaining 3 cases, one had the right group assignment after correcting superimposition problems and for the two remaining the clinical experts agreed the computer decision was more accurate.
Conclusions: Our work demonstrated that it is possible to classify OA pathology using 3D Shape Analysis of the TMJ. The knowledge gained from this project will allow more rational clinical decision making for TMJ OA patients.
Methods: The study sample consisted of 91 healthy and TMJ OA patients. Our approach utilized Statistical Shape Modeling (SSM) to obtain information about TMJ OA phenotypes. All healthy subjects had radiographic diagnosis of no condylar pathology; TMJ OA patients underwent a clinical exam by an orofacial pain specialist to confirm the TMD research diagnostic criteria. All participants had a CBCT scan taken (i-CAT Next Generation, 120 kV, 18.66 mA). 3D models of the TMJs were generated from CBCT, and left joints were mirrored into the right side. Correspondent Point distribution models (PDM) were constructed using SPHARM-PDM. After quality control, the cohort comprised 218 TMJ joints (153 TMJ OA, 65 healthy). TMJ OA joints were by consensus subdivided into 7 subgroups based in pathological morphological phenotypes. We generated SSM-shape spaces to describe the morphology of each TMJ OA subgroup and the healthy group. Then we projected each TMJ OA case into each SSM-shape space, and shape loads were computed to calculate similarity with that space. A case will be classified in the group for which the similarity metric is smaller.
Results: 9.8% TMJ OA cases were misclassified. No TMJ OA case was classified in the healthy group. From the 15 out of 218 misclassifications, 12 had the correct classification as second closest. For the remaining 3 cases, one had the right group assignment after correcting superimposition problems and for the two remaining the clinical experts agreed the computer decision was more accurate.
Conclusions: Our work demonstrated that it is possible to classify OA pathology using 3D Shape Analysis of the TMJ. The knowledge gained from this project will allow more rational clinical decision making for TMJ OA patients.