Periodontitis Risk Assessment using two artificial Neural Networks-A Pilot Study

Abstract

Artificial neural networks are currently used for a variety of complex problem solving approaches where a conventional method may not be feasible. Aims & objectives: Levenberg Marquardt and Scaled Conjugate Gradient feed forward back propagation neural network algorithms were compared to assess accuracy for periodontitis risk prediction. Materials and Method: In the present study 230 subjects were assessed for major and minor periodontitis risk factors such as; age, gender, family history of periodontitis, history of periodontal surgery, diet, smoking history, pan chewing habit, history of diabetes, history of hypertension, presence of sub gingival restorations, bleeding on probing, debris index (OHI-S), average pocket probing depth, presence of root calculus, presence of furcation involvement and vertical bone loss. Periodontitis risk assessment was done on a grade of 1 to 5. Results: The Levenberg Marquardt algorithm performed considerably better than Scaled Conjugate Gradient algorithm by converging faster with lesser iterations and produced minimum mean square error in both training and simulation phases. Conclusion: A properly trained neural network with Levenberg Marquardt back propagation algorithm can effectively be used for periodontitis risk prediction.