Use of an artificial neural network for the diagnosis of myocardial infarction

Article Abstract:

Proper diagnosis of illness requires consideration of a large quantity of patient data, a myriad of diagnostic possibilities, and an immense accumulation of information about human disorders. Although it might appear to be a problem for which computers might find application, computer-aided diagnosis has gained only little use. In general, diagnostic computers are time-consuming, difficult to use, and have not proved to be superior to the more traditional methods used by physicians. Neural networks may provide a new avenue for development of computerized diagnostic tools. Neural networks are programs which work not by a strict set of rules, but rather by maintaining a complex array of interconnecting nodes which are altered in response to input, or ''experience.'' Such neural networks have proved to be superior for some applications, particularly applications involving the recognition of patterns. A neural network program has now been designed to aid in the diagnosis of myocardial infarction, the most common form of heart attack. Proper diagnosis of heart problems is critical, since the appropriate treatments are strongly dependent upon the underlying cause. Furthermore, rapid diagnosis is also critical, since the effectiveness of some treatments depends upon rapid initiation. The neural net was trained on the case records of 351 hospitalized patients who were deemed to be suffering a myocardial infarction. After the training of the neural net, the ability of this program to recognize the features characteristic of a heart attack was evaluated on 331 consecutive patients presenting with chest pain. Physicians diagnosing these cases were found to have a sensitivity of 78 percent; that is, 78 percent of the actual myocardial infarctions were recognized as such. The specificity achieved by the physicians, that is, the fraction of cases which were not myocardial infarction and were so diagnosed, was 85 percent. The neural network achieved sensitivity and specificity of 97 and 96 percent, respectively, a result superior to that of the physicians. Of course, the reliability of a neural network for the diagnosis of myocardial infarction, or any other disorder, will require the evaluation of many more cases under carefully controlled conditions. However, the present study demonstrates the feasibility of neural networks as a useful diagnostic aid. (Consumer Summary produced by Reliance Medical Information, Inc.)

---

Neural networks: what are they?

Article Abstract:

In the December 1, 1991 issue of the Annals of Internal Medicine, a physician presents the development of a neural network computer program for the diagnosis of heart attacks. When tested on a series of emergency room patients, this neural net performed somewhat better than physicians. Unlike most computer programs, neural networks operate by varying the values assigned to nodes and interconnections of a large network modelled in the computer's memory. These values change with experience, and the output of the neural network gradually becomes more accurate as its ''experience'' increases. Such programs have proved capable of many tasks involving pattern recognition, tasks which have proved extremely difficult for more conventional programming techniques. However, one consequence of the architecture of the neural network is that the relationship between the input data and the output diagnosis is obscure. Just as a child may readily tell a cat from a dog, but be unable to explain the difference in a satisfactory way, a neural network cannot explain why the diagnosis of myocardial infarction was given, even if it is indeed correct. Are physicians likely to use such a mysterious device? Physicians should already be familiar with many cases in which the empiric evidence that something works is more important than knowing why. The action of some drugs is poorly understood, but large amounts of data indicate that they should be effective in certain cases; physicians have no qualms about using them in these cases. Neural networks should be treated in the same way. The must be evaluated in a reliable, objective, and unbiased manner. Their use should be determined on the basis of experimental data which support their value, or not. (Consumer Summary produced by Reliance Medical Information, Inc.)

Editorial

---

Perioperative assessment and management of risk from coronary artery disease

Article Abstract:

The relative value of various tests used to evaluate risk for heart complications among prospective elective surgical patients is presented. Patients identified as low-risk or intermediate-risk using baseline assessment tools likely do not need any further testing before non-vessel surgery. Imaging or stress tests may be valuable in further identifying risk among intermediate-risk patients before vessel surgery. High-risk patients may require preliminary corrective surgery, treatment with beta-blocking medication, or counseling about canceling elective surgery.

Prevention, Complications and side effects, Surgery, Coronary heart disease, Postoperative complications, Cardiac patients

Similar abstracts:

• Abstracts: Coronary angiography after thrombolytic therapy for acute myocardial infarction. Evidence-based coronary care
• Abstracts: Reducing mortality in patients with extensive myocardial infarction. Thrombolysis in acute myocardial infarction
• Abstracts: Adenosine and supraventricular tachycardia. A clinical trial of the angiotensin-converting-enzyme inhibitor trandolapril in patients with left ventricular dysfunction after myocardial infarction
• Abstracts: Body MR imaging and the local staging of neoplasms. Trigeminal neuralgia: MR imaging features
• Abstracts: The Patient Self-Determination Act and the future of advance directives. part 2 Racial differences pertaining to a belief about lung cancer surgery: results of a multicenter survey

This website is not affiliated with document authors or copyright owners. This page is provided for informational purposes only. Unintentional errors are possible.