Ischemia detection with a self-organizing map supplemented by supervised learning

The problem of maximizing the performance of the detection of ischemia epis odes is a difficult pattern classification problem. The state space for thi s problem consists of regions that lie near class separation boundaries and require the construction of complex discriminants while for the rest regio ns the classification task is significantly simpler. The motivation for dev eloping the supervising network self-organizing map (sNet-SOM) model is to exploit this fact for designing computationally effective solutions both fo r the particular ischemic detection problem and for other applications that share similar characteristics. Specifically( the sNet-SOM utilizes unsuper vised learning for the "simple" regions and supervised for the "difficult" ones in a two stage learning process. The unsupervised learning approach ex tends and adapts the self-organizing map (SOM) algorithm of Kohonen, The ba sic SOM is modified with a dynamic expansion process controlled with an ent ropy based criterion that allows the adaptive formation of the proper SOM s tructure. This extension proceeds until the total number of training patter ns that are mapped to neurons with high entropy land therefore with ambiguo us classification) reduces to a size manageable numerically with a capable supervised model, The second learning phase (the supervised training) has t he objective of constructing better decision boundaries at the ambiguous re gions. At this phase, a special supervised network is trained for the compu tationally reduced task of performing the classification at the ambiguous r egions only. The utilization of sNet-SOM with supervised learning based on the radial basis functions and support vector machines has resulted in an i mproved accuracy of ischemia detection especially in the last case. The hig hly disciplined design of the generalization performance of the support vec tor mai chine allows designing the proper model for the number of patterns transferred to the supervised expert.