Fracture Feature Detection in Dental Ceramics Using Image Segmentation

Objectives: Fracture toughness is predictive of the longevity of dental prostheses. Measuring fracture toughness is important in ceramics as they are susceptible to intrinsic flaws, such as defects and pores, resulting in variable strength. Surface crack in flexure (SCF) is a standard technique for measuring fracture toughness for all ceramics irrespective of the microstructure, including the newer generation of graded zirconia. Although the SCF technique does not require sophisticated equipment, it does require discerning the boundary of the critical flaw, which is time-consuming and requires expertise. The objective of this study is to segment the fracture surface images of one glass and one glass-ceramic to develop and train a convolutional neural network model for identifying characteristic fracture features.
Methods: Scanning electron micrographs of previously analyzed fracture surfaces of fused silica glass (Viosil) and nanofluorapatite glass-ceramic (IPS e.max Ceram) were gathered. These rectangular beam specimens (3x4x28mm³) were fractured using the SCF technique as per ASTM C1421. A series of image pre-processing steps were performed including cropping, brightness and contrast adjustment, and median filtering. The processed images were segmented using the Canny Edge detector to identify the characteristic features on the fracture surface.
Results: The Canny edge detector was successful in detecting the boundary of the critical flaw for materials with two different microstructures (glass and glass-ceramic). The difference in the fracture toughness calculated based on flaw boundary detected by experienced fractographers vs. segmented images was 1.24%. Additional fractographic features such as wake hackle and twist hackle were also identified. These features can provide information about the reasons for failure in clinically failed dental restoration.
Conclusions: Characteristic fracture features (including the boundaries of the critical flaws) were identified using image segmentation algorithms. These segmented images would be fed into a neural network architecture to train it to accurately identify the fractographic markings and compute the fracture toughness of the ceramic.