Genetic Algorithm for Design Optimization of Dental Implants

Objectives: We previously reported on cyclic loading of 4 commercially available reduced-diameter implant systems to determine which design features are associated with fatigue resistance (Loeb et al., 2014), using finite element modeling software to predict the fatigue limits of reduced-diameter implant designs (Satpathy et al., 2021), training an artificial neural network (ANN) to predict the fatigue limits of implants (Satpathy et al., 2022), and guiding the ANN towards an optimal design using a Latin hypercube to generate points in the 16-dimensional design space vs manual searching (Griggs et al., 2023). The current study was aimed at more efficient design optimization by using genetic algorithm method and ANN in tandem.
Methods: The 16 design parameters of implant designs were coded as 16 genes in a virtual genome. Successive generations of designs were created by random crossover between the parent genomes with 125 offspring per generation, point mutation rates ranging from 1% to 20%, and 2 to 10 parents per generation. The previously trained ANN was queried regarding the fitness (fatigue limit) of each offspring, and the fittest offspring were chosen to be parents of the next generation. Design parameters were constrained to be within either 20% or 40% of the commercially available implant (Biomet 3i external hex).
Results: Regardless of the genetic algorithm parameters chosen, the implant designs rapidly evolved with the fatigue limit reaching 264 N, which is 128% higher than the best of the commercially available products that we have tested (116 N). This also exceeded the performance of designs found by Latin hypercube (228 N) and manual search (254 N). The speed of convergence on the optimal design was directly related to the point mutation rate and was independent of the number of parents per generation.
Conclusions: Genetic algorithm method is more efficient than Latin hypercube and manual search in optimizing the fatigue limit of reduced-diameter dental implants. However, these results still need to be validated by cyclic loading of physical prototypes.