Quantitative analysis of the regional acceleratory phenomenon produced by various bone biomodification techniques using deep learning
Di Battista, Massimo
MetadataShow full item record
The Regional Acceleratory Phenomenon (RAP), a post-injury transient bone remodeling phenomenon, is the foundation of most modern corticotomy-assisted orthodontics treatment. The piezoelectric knife (PIEZO) is an alternative to mechanical devices in conventional oral surgery procedures by rotary bur (BUR). It enables cutting bone with low ultrasonic frequency while protecting fragile anatomical structures. PIEZO may enhance RAP leading to cellular activities while BUR causes severe trauma in the medullary tissue resulting in excessive inflammation. Our previous PIEZO study demonstrated that the trans-cortical penetration (TCP) significantly activated biological responses by RAP more than the intra-cortical defect (ICD). PIEZO-TCP increases initial osteocyte apoptosis, osteoclast, and osteogenic activities. The cortical alveolar bone is a crucial structural element to support teeth or implant stability. Yet, there is no reliable metric for predicting the mechanical properties of the bone in this critical region. Recent studies suggest that micro-porosity assessed by deep learning from micro-CT images correlates with cortical bone’s elastic modulus and ultimate compressive strength by tissue mineral density. We hypothesized that cortical porosity might increase in the resorption phase and decrease in the formation phase associated with RAP. In this study, we used deep learning analysis to compare cortical micro-porosity from post-operative micro-CT images of PIEZO, BUR, and Control at day 7 and 14. Eighteen 9-week-old male Sprague-Dawley rats were randomly divided into three groups: PIEZO, BUR and Control with deep and shallow defects on the right and left tibias in test groups. 3D rendered micro-CT images, with approximately 1000 slices each, were analyzed for cortical micro-porosity with deep learning algorithms for multi-label segmentation. The deep learning model was trained to analyze the image and classify the pixels in one of these labels: background, cortical bone, reactive calcified tissue and cortical porosity. Cortical porosity was considered to be all void in the cortical bone, except for the notch or canal of the tibial nutrient artery. Regional cortical porosity was assessed using the full scan volume (approximately 6mm long). Local cortical porosity was measured and compared for 3 standardized local regions of interest (ROIs, 0.5, 1.0, 1.5mm from the defect edge). Results indicate that Piezo corticotomies have a significantly deeper impact on the RAP versus conventional rotary burs and that the deep learning process, a subset of machine learning that makes the computation of multi-layer neural networks, could be a very powerful new tool and an innovative approach to dental research.