A Connectionist Aid to the Early Diagnosis of Myocardial Infarction

Abstract

The development of a decision aid for the early diagnosis of myocardial infarction is described. The system uses a connectionist system or neural network called the multilayer Perceptron. It is shown that the performance of the network can exceed that of the admitting clinicians, a panel of senior physicians in a large teaching hospital and a protocol derived using conventional statistical methods, over a wide range of performance measures. In particular, the network demonstrates the highly specific behaviour necessary when making the decision whether or not to administer thrombolytic therapy. This therapy reduces mortality but only if given in the few hours after the onset of symptoms and long before confirmatory laboratory test results are available.

The network is compact and has been implemented on a portable computer. In operation it responds very quickly, giving its diagnosis and recommendations (taking account of clinical opinion) in a fraction of the time taken to input the patient’s symptoms. An additional novel feature of the system is the use of “contribution analysis” to give an indication of the relative importance of individual data items in the decision making process.