PubMed Health⌕ Search

Biomedical subjects

J L Talmon

Publications and source records attributed to J L Talmon.

36 records · Page 2Linked to original sources

On the quality of neural net classifiers.

This paper describes several concepts and metrics that may be used to assess various aspects of the quality of neural net classifiers. Each concept describes a property that may be taken into account by both designers and users of neural net classifiers when assessing their utility. Besides metrics for assessment of the correctness of classifiers we also introduce metrics that address certain aspects of the misclassifications. We show the applicability of the introduced quality concepts for selection among several neural net classifiers in the domain of thyroid disorders.

Humans↗

Artificial neural networks and ECG interpretation. Use and abuse.

The computerized interpretation of the resting electrocardiogram has reached a steady-state phase: an equilibrium between sensitivity and specificity has been reached. New computer techniques, such as expert systems and artificial neural network technology, have been proposed or are currently under evaluation. Although neural network techniques are based on complex mathematical theories and their application is full of pitfalls, progress has been made in a number of subdomains, like signal filtering, electrocardiographic classification, and compression of stress electrocardiograms. Presently, the hesitating acceptance by the human user forms one of the obstacles that needs to be overcome by convincing, well-performed studies.

Diagnosis, Computer-Assisted↗

The effect of noise and biases on the performance of machine learning algorithms.

This paper describes the results of experiments with a machine learning algorithm for the induction of classification trees. We mainly address the impact of noise on the resulting classification tree and on the classification results obtained with the derived tree. We use the domain of the biochemical assessment of thyroid diseases as an example. Some suggestions for quality assessment are outlined that should be available in tools that assist users in deriving classification trees in noisy domains.

Algorithms↗

Evaluation of computerised questionnaires designed for patients referred for gastrointestinal endoscopy.

The aim of this prospective study was to assess the applicability of a computerised medical history system at the open-access endoscopy unit of a University Hospital during routine clinical practice. We studied feasibility, acceptability and reproducibility of computerised questionnaires designed for patients referred for endoscopy of either the upper or lower gastrointestinal (GI) tract, partly in comparison with almost identical paper questionnaires. In the first period of the study 1134 patients were referred of whom 73% answered the paper questionnaire, and during the second period 537 patients were referred of whom 64% answered the computerised questionnaire (P less than 0.001). There was no sex dependency regarding the ability to answer both types of questionnaires. A significant age dependency regarding the ability to answer the paper questionnaires was observed for patients referred for upper and lower GI endoscopy (P less than 0.0001 and P less than 0.0001). Corresponding with this observation, the ability to answer the computerised questionnaires significantly decreased with increasing age for both groups of patients (P less than 0.0001 and P less than 0.0001). The average completion times of both computerised questionnaires, designed for the upper and lower GI tract, were 11 min. The completion times did not depend on sex, but there was a slight positive correlation with age (P less than 0.0001 with multiple R = 0.27 and P = 0.0593 with multiple R = 0.12). Answering our computerised questionnaires was consistently reported to be interesting and easy, and did not last too long according to the respondents. After patients had answered the computerised history system, the insight into their complaints had increased in 40%, and 45% could put their complaints into words more easily. Only 25% of the patients were unable to express all their complaints. A reproducibility of medical history data of 85% was observed between paper and computerised questionnaires. Our computerised questionnaires can provide us with a detailed and uniformly acquired medical history of most patients referred for GI endoscopy. They are well accepted by the patients and can be completed during the normal waiting period. The medical history data obtained with these questionnaires are quite reproducible.

Adult↗

Electrocardiogram waveform in fetal sheep: a system for ECG analysis.

Computerised ECG analysis is helpful for the analysis of fetal ECGs, recorded during animal experiments. It provides a tool for rapid measurement of characteristics of ECG signals. Such an ECG analysis system is described. The different steps of the signal analysis are discussed and preliminary results are presented. The current version of the program is implemented in Turbo-Pascal and runs on an IBM-compatible PC with an 8087 processor and an EGA colour display.

Animals↗

Multilead ECG analysis.

This paper describes the results of our recent research in computer-assisted ECG/VCG interpretation. It comprises new developments which were initiated by the advent of relatively inexpensive microcomputers. Our previous systems performed an off-line analysis of ECGs. Currently, there is a trend to move computer power near to the patient and to provide on-line analysis of ECGs. Besides the advantage of the direct availability of the ECG interpretation, quality control will reduce the number of uninterpretable ECGs and hence the number of repeated recordings. This paper describes the requirements that were established for a system for on-line ECG analysis. The system is based on our modular approach, just like our off-line system, Modular ECG ANalysis System (MEANS). Changes in the methods and software had to be made mainly because of the simultaneity of all ECG leads and the concurrency of the processing tasks. Other modifications and extensions of the algorithms necessary to meet the requirements of on-line ECG interpretation especially those related to processing speed, are discussed, and evaluation results are presented.

Electrocardiography↗

Algorithms for the detection of events in electrocardiograms.

The detection of events in electrocardiograms (ECGs) and vectorcardiograms (VCGs) is an essential processing step in systems for computer-assisted ECG/VCG interpretation. This paper describes various algorithms for the detection of QRS complexes, atrial activity and artefacts. These algorithms have been used in various versions of the TNO Modular ECG/VCG Interpretation System including its latest version, MEANS. Evaluation results are presented.

Biometry↗

A computer program for the analysis of serial electrocardiograms from patients who suffered a myocardial infarction.

A program is described, which compares serial ECG recordings of patients, who suffered a myocardial infarction. The comparison of the ECGs is based on the measurements, delivered by the Modular ECG/VCG Interpretation System. The two most recent ECGs are compared and a trend analysis based on all ECG recordings of the patient is performed. The criteria in the comparison program are based, among others, on data obtained from a combined cardiological and respiratory survey in The Netherlands in 1973 and 1976.

Computers↗

Clinical significance of upsloping ST segments in exercise electrocardiography.

Conventional exercise electrocardiographic criteria usually involve patterns with a horizontal or downsloping ST segment. In the present study we present criteria based on upsloping ST segments and compared these criteria with the conventional criteria. Using upsloping ST-segment criteria, the amount of ST-segment depression at 80 msec after the end of the QRS complex is used as a parameter (ST criterion E, with a depression of 100 mV, and ST criterion F, with a depression of 200 mV). In the graded exercise test a bicycle ergometer was used. The ECG leads were CM5 and CC5. The results of exercise electrocardiography were compared with the findings from coronary arteriography. In 623 selected patients (565 males and 58 females), application of conventional ST criteria gave a sensitivity of 56% and a specificity of 94%; with application of the ST criteria E or F, sensitivity was 75% and specificity 90%. In the 58 females use of these new criteria resulted in a sensitivity of 76% and specificity of 88%. Ninety-three patients (15%) could be classified as positive exercise responders by the sole presence of an upsloping ST segment (type E or F). Sixty-eight percent of the patients with type E and 75% with type F had two- or three-vessel disease (coronary obstructions greater their or equal to 50%). We conclude that ST criteria based on upsloping ST segments significantly increase the diagnostic yield of the exercise ECG.

Coronary Angiography↗

A computer program for ECG classification according to the Minnesota code.

A program for the automated generation of Minnesota-codes from the standard 12-lead electrocardiogram (ECG) recorded at rest has been developed. The program has been written as a part of the Modular TNO ECG/VCG Processing System. Results of the program on two sets of ECG'S (279 predominantly normal recordings and 286 consecutive recordings from a Cardiological Department) are presented. These electrocardiograms were also coded by hand by different ECG technicians. An agreement of 83% between hand and computer coding was obtained. One-third of the disagreements (6% of the cases) is caused by small differences (less than 5%) in the measurements between technicians and computer.

Diagnosis, Computer-Assisted↗

The advantage of modular software design in computerized ECG analysis.

In the early 1970s, modular software design was advocated as a means to obtain better structured and more reliable software systems. Such a modular approach was also introduced for computerized ECG processing in 1972. The TNO Modular ECG/VCG interpretation system was developed along these lines. During the last decade, the algorithms of this system were thoroughly improved and evaluated. This resulted in 1983 in a new version of this system named MEANS (Modular ECG ANalysis System). This paper outlines the ideas of the modular approach as they were originally defined in 1972 and developed over the years. The usefulness and advantages of these ideas will be discussed based on 12 years of experience. It appeared that modularity is the ideal approach to the development and maintenance of complex systems such as for ECG/VCG interpretation.

Computers↗