Classification Methods Using Neural Networks and Partial Precedence Algorithms for Differential Medical Diagnosis: A Case Study

The problem of correctly diagnosing different types of ailments has been tackled with different artificial intelligence techniques since its inception. Both heuristic and statistically based algorithms have been discussed in the past. In this paper we establish a comparison between one heuristic algorithm based on partial precedence and majority decision rules and two types of statistical ones: multi-layer perceptrons (MLP) and self-organizing maps (SOMs) when applied to the automated diagnosis and treatment of cleft lip and palate. We show that although all three methods perform reasonably well (with efficiency ratios better than 0.9) the neural networks achieve their goals with a considerably diminished set of data without detriment in their performance. Furthermore, we are able to tackle an enlarged set and still retain the high yields with the use of MLPs and SOMs.