COMPUTATION OF DIFFERENTIAL DIAGNOSIS IN ELECTROCARDIOGRAPHY *

The primary goal of any electrocardiographic analysis is the establishment of a differential diagnosis for each individual case. In automatic computational procedures, the output should preferably consist of a listing of all possible ECG diagnoses with a probability value for each diagnostic entity. The automatically computed differential diagnosis of the electrocardiogram should serve as a building stone in the larger framework of a final clinical diagnosis. Evaluation of the ECG in the context of the patient's history, physical. and laboratory findings is still left to the judgment of the physician who has to be aware of the natural limitations of ECG information. It has always to be remembered that the electrocardiogram provides primarily information on the morphology of the heart and only to a limited extent on its function. With these limitations in mind, it appears desirable to provide the physician with a quick means for obtaining the maximum information which can be extracted from the electrocardiogram. Large-scale digital computers appear well suited for attempting this task. Any type of ECG analysis can be performed automatically and the complexity of measurements are not a limiting factor. Classification of records in various pathological groups can be derived from statistical probabilities based on large clinical data collections and autopsy material. Thus, subjective data evaluations can be replaced by quantitative statements leading to the desired diagnostic classifications. Attempts to achieve the described goals have been in progress in the Veterans Administration for several years. Technical requirements for automatic ECG analysis by digital computer have been described in detail elsewhere.' 4 Only a few essential considerations shall be mentioned here briefly. An important first step represents the selection of proper ECG leads. Since it had been shown that conventional bipolar and unipolar leads differ consid. erably from one subject to another in effective direction and lead strength," more reliable leads had to be selected in the form of corrected orthogonal leads. Such a choice is important for two reasons. First, information collected