Is computerized diagnosis possible?

Computer-based clinical decision aids. A review of methods and assessment of systems

During the last three decades a great deal of research has been devoted to the development of integrated clinical decision support systems. This report aims to give a basic understanding of what is required for such a system. By means of a large literature study a survey is given of the major components of computer-based clinical aid systems. The main approaches and several aspects of evaluation of such programs are described. The computer has several inherent capabilities which are suitable for medical problem solving and can help in the formalization of medical knowledge. The components of such systems include the computer database, the reasoning engine and the user interface. The different approaches on which the reasoning engine is built are based on manipulation of information and advocate the use of knowledge to construct a solution to a problem. The information in the mode vary from data-intensive to knowledge-intensive. Assessment of decision support systems is a very important phase in the development of such systems. Evaluation should be made on the accuracy of the program, the nature of the system, the use of the data and the acceptance by the target users. Whatever the model is, its effectiveness will depend on the data with which the program has to work. Acceptance by physicians depends among other things on ease of use of the user interface. Profound changes in the delivery of health care will be induced through the rapid growth of on-line computer communication together with the development of integrated clinical decision support systems and electronic medical records. Notwithstanding the rapid growth of computer technology, computer-aided decision making is in its infancy and real support in daily practice is not yet achieved.

Comparison of three arterial pulse waveform classification techniques

Peripheral pulse waveforms can become stretched and damped with increasing severity of peripheral vascular disease (PVD) and hence could provide valuable diagnostic information. This study compares the diagnostic performance of 3 established classification techniques (a linear discriminant classifier, a k-nearest neighbour classifier, and an artificial neural network) for the detection of lower limb arterial disease from pulse waveforms obtained using photoelectric plethysmography (PPG). Pulse waveforms and pre- and post-exercise Doppler ultrasound ankle to brachial pressure indices (ABPI) were obtained from patients attending a vascular measurement laboratory. A single PPG pulse from each big toe was recorded direct to computer, pre-processed, and then used as classifier input data. The correct classifier outputs were the corresponding ABPI diagnostic classification. Pulse and ABPI measurements from 100 legs were used as training data for each classifier, and the computed classifications for pulses from a further 266 legs were then compared with their ABPI diagnoses. The diagnostic accuracy of the artificial neural network (80%; was higher than for the optimized k-nearest neighbour classifier (k = 27, accuracy 76% and the linear discriminant classifier (71%). The Kappa measure of agreement which excludes chance was highest for the artificial neural network (57%) and significantly higher than that of the linear discriminant classifier (Kappa 40%, p < 0.05). The value of Kappa for the optimized k-nearest neighbour classifier (k = 27) was intermediate at 47%. This study has shown that classifiers can be taught to discriminate between small, and perhaps subtle, differences in features. We have demonstrated that artificial neural networks can be used to classify arterial pulse waveforms, and can perform better overall than k-nearest neighbour or linear discriminant classifiers for this application.

Computer-assisted diagnosis in gastroenterology

During the last 20 years, Danish gastroenterologists and biostatisticians have been involved in practical and theoretical research aimed at providing computer-assisted decision support in clinical practice. On the practical side, we summarize two clinical applications, one in the acute abdomen, the other in jaundice. The former project is part of a worldwide effort which appears able to reduce considerably the perforated appendix rate and the negative laparotomy rate. In the latter project, also entwined in international activities, three methods of early statistical discrimination of jaundice causes have been tested, and it has been shown how ultrasonographic data can be incorporated into the diagnostic assessment. On the theoretical side, a Dutch-Danish group has been looking into how one best designs a trial and analyses its data in the context of measuring the quality and impact of machine-made diagnostic advice. Having pointed out the international ramifications of these activities, we outline what may lie ahead: Gastroenterologists must be prepared to join large-scale field trials of clinical advice-giving software. In the future, however, such software will often become embedded in general hospital recordkeeping systems.

Medical diagnostic decision support systems--past, present, and future: a threaded bibliography and brief commentary

Articles about medical diagnostic decision support (MDDS) systems often begin with a disclaimer such as, "despite many years of research and millions of dollars of expenditures on medical diagnostic systems, none is in widespread use at the present time." While this statement remains true in the sense that no single diagnostic system is in widespread use, it is misleading with regard to the state of the art of these systems. Diagnostic systems, many simple and some complex, are now ubiquitous, and research on MDDS systems is growing. The nature of MDDS systems has diversified over time. The prospects for adoption of large-scale diagnostic systems are better now than ever before, due to enthusiasm for implementation of the electronic medical record in academic, commercial, and primary care settings. Diagnostic decision support systems have become an established component of medical technology. This paper provides a review and a threaded bibliography for some of the important work on MDDS systems over the years from 1954 to 1993.

Computer graphics representation of a statistical model used with computer-aided diagnosis

A description of computer graphics of a multidimensional model that is used with computer-aided diagnosis or prognosis is presented. The model is discussed and computer graphics of the model are developed. The computer graphics are suitable as visual supplements for presenting the computer-aided diagnostic model to individuals who may be inexperienced in multivariate statistics.

Specialized PC software package for creation of computer systems supporting partial diagnostics based on numerical results of medical examinations

Computer scientists creating computer systems supporting partial diagnostics based on numerical results of medical examinations always follow a similar method. Why not algorithmize this procedure? This was the main reason for preparation of the specialized PC software presented in the paper. The software is addressed to clinicians-scientists and enables such users to create by themselves (with no need of programming) practically useful computer systems for differential diagnosis of chosen diseases. These generated systems are assigned to an end-user, i.e. any physician interested in a partial diagnosis of the considered type. They perform classification of patients on the basis of numerical results of the examinations selected for the training databases. The software is based on a statistical approach; multi-dimensional variance analysis and discriminant analysis methods have been used. The pilot version of the software (experimentally implemented in a Warsaw clinic) and an outline of planned work are presented.

A menu-driven knowledge base browsing tool

Conventional computer-assisted medical decision-making systems have had limited impact on routine clinical practice. This has stimulated an alternative approach to the utilization of medical knowledge bases. Centering on the storage and retrieval of medical information, it aims to provide clinicians with computerized medical reference systems. In this paper we describe the development of a prototype menu-driven browsing tool, which allows clinicians to browse through the contents of a knowledge base in a number of ways. Operations include interrogation via disease classes, names or attributes; hierarchical display of all or part of a disease profile; printing of a disease profile; construction of differential diagnosis lists and comparison of two diseases. We discuss how the use of a menu-driven interface can help to overcome some of the problems encountered with previous designs of medical reference systems.

Use of computers in diagnosis of jaundice

The development of computer-assisted diagnostic systems for the differential diagnosis of jaundice has been attempted in several studies during the last 25-30 years. All working systems have depended on numerical methods whereas expert systems have not yet become operational. The first step in the construction of a system is the collection of a data base from a series of jaundiced patients of clinical information (indicants) i.e. symptoms, signs, and the results from laboratory tests. The best discriminating indicants are selected and processed into a mathematical rule. The performance of this rule must be tested on an independent test sample of relevant patients. The performance may also be compared to that of clinicians. So far the computerized diagnoses have not been more accurate than those of clinicians. However, computer-assisted diagnostics may form an important first step in clinical decision making regarding the selection of a confirmatory diagnostic test (direct cholangiography, ultrasonography, liver biopsy, etc.) in the evaluation of the jaundiced patient.

Computer-aided diagnosis: a review

Fifteen years ago, computer-aided diagnosis of the acute abdomen promised much. Today it is little used. Studies have been flowed by poor trial design, bias, selective reporting of results, statistical naivety and spurious conclusions. The computer system lacks 'common sense' and is less accurate than clinicians. Yet its introduction has been associated with improved patient management and outcome. Much of the effect arises from structured data collection methods and some from audit feedback to clinicians. It is on these innovations, not on the computer, that future work should focus.

Computerized dermatopathologic diagnosis

A practical, transferable microcomputer system for differential diagnosis in dermatopathology, called TEGUMENT, has been developed for use by dermatologists on the standard IBM PC, Compaq, and other compatible personal microcomputers. In an interactive computer program a set of information is abstracted from the microscopic study of each specimen by a dermatologist, to compare with a structured knowledge base. The process leads through a relevant sequence of descriptive phrases until the findings can be allocated to a disease class. The microscopic description and diagnosis are then combined with clinical information by the computer and printed, optionally, as a pathology report. The identification and diagnosis of each case are preserved in permanent memory to enable future search and sorting. The results of independent validation are that a pathologist made the same diagnosis as the machine or a similar differential diagnosis in 91.8%, disagreed in 4.8%, and was unable to make a diagnosis from the description furnished by the machine in 3.4% of 147 actual cases. We conclude that a certain critical minimum of information is required for objective diagnosis; more information is needed for definitive than for differential diagnosis; a characteristic feature is necessary to distinguish between differential diagnoses; an objective description may admit of more than one diagnosis; ambiguity may be reduced by presenting for consideration all distinguishing features that characterize closely related diagnoses; and the personal microcomputer, programmed in this way, is of considerable assistance to the dermatologist in the histopathologic diagnosis of diseases and neoplasms of the skin.

Evaluating medical expert systems

Approximately 90% of all computerized medical expert systems have not been evaluated in clinical environments. This paper: identifies the principal methods used to assess the performance of medical expert systems in both laboratory and clinical settings, describes the different research strategies used in the evaluation of medical expert systems at different development stages, and discusses past evaluation efforts in relationship to future applications of different decision support technologies and expert systems in health care.

Computer diagnosis in jaundice. Bayes' rule founded on 1002 consecutive cases

Extensive clinical and clinical chemical information was collected from 1002 consecutive jaundiced patients. Initial selection of variables based on Chi 2-tests or Mann-Whitney U-test allowed the removal of 64 of the 107 variables originally collected. A further selection of variables was carried out using a modified version of Bayes' rule thus reducing the number of variables from 43 to 22. Of the 982 patients with a final diagnosis 743 patients (76%) could be classified correctly into one of 13 diagnostic categories. The Bayes' rule was also applied to a test group of a further 110 jaundiced patients and found to perform equally well: of 108 patients with a final diagnosis 81 (75%) were correctly classified. A comparison between the clinician's diagnosis and the computer-aided diagnosis according to Bayes' rule demonstrated agreement with regard to one of the 13 diagnostic alternatives in 734 patients (75%), of whom 81 patients were wrongly diagnosed. In the test group agreement upon diagnosis was found in 80 patients (74%). By plausibly combining the computer-aided and the clinician's preliminary diagnoses, more correct classifications were obtained than with either method alone. Many diagnostic modalities such as ultrasound examination, CT-scan, and direct cholangiography are at hand today for the differential diagnosis of jaundice. Computer-aided diagnosis using Bayes' rule has proved a reliable tool for the clinician and can be used in the planning of a diagnostic strategy for the individual jaundiced patient.

A survey of decision aids for clinicians

Inconsistency in applying medical knowledge is a major reason for varying standards of medical care. Five types of aid have been introduced into medicine to help decision-making: questionnaires, algorithms, database systems, diagnostic systems, and, finally, computer-based decision-support systems. Of these, the most effective act as reminder or prompt systems to assist doctors without threatening their clinical freedom.

Differential diagnosis of liver parenchymal diseases by likelihood method using 12 laboratory data and age

Liver parenchymal diseases were statistically diagnosed by likelihood method using 12 routine liver function tests and age. 444 cases of liver diseases were classified into 8 groups by histological diagnosis. A score diagnosis table was made from the data of these cases. For the likelihood diagnosis, data of each case were adapted to the score table and the probable diagnosis was calculated. Correct diagnosis rate of the first probable diagnosis was 50% in all cases and that of the first and the second was 71%. Descending order of the correct diagnosis rate of the first diagnosis was fatty liver (76%), liver cirrhosis (67%), slight histological changes (61%), acute hepatitis (51%), alcoholic liver injury (48%), chronic aggressive hepatitis 2A (43%), chronic persistent hepatitis (40%) and chronic aggressive hepatitis 2B (26%). In conclusion, differential diagnosis of liver parenchymal diseases was made easily with the score table of 13 informations with a considerable success.

Training in medical informatics. The use of computers for diagnostic purposes

Part of a blockcourse on medical informatics is presented; this course is intended for medical students. It is shown how medical students are introduced to the study of the role of computers for diagnostic purposes. The course consists of an oral presentation which introduces the student to the subject, and of practical work on systems in order to more fully comprehend the topics explained in the oral part of the blockcourse. In the oral presentation the student is introduced to various concepts that are used in computer-aided diagnosis. A critical review of the possibilities of computer use for diagnostic purposes is given. A system is presented, with which the student can work interactively. It consists of a database of patients, referred to the hospital because of suspected congenital heart disorder. Bayes' rule and diagnostic tree decision schemes are available to the student to acquaint himself with the subject. The ways he can work with this system are explained. The course is given regularly (every 4 months) to medical students and is well appreciated.

Applying the principle of logical elimination to probabilistic diagnosis

A probability theorem is described which identifies the factors to be considered when applying the principle of logical elimination to probabilistic diagnosis. The proof is based only on the multiplication, addition and universal bound axioms of probability. When new tests are being designed from knowledge of disease mechanisms, criteria based on the theorem can be used to assess their suitability for interpretation in a logical manner. A simple convention is adopted to allow probabilities to be substantiated in terms of observed frequencies.

Computer assisted diagnosis and computer consultation in neurology: preliminary testing of diagnostic accuracy for the neurologist system

The Neurologist system is a computer program for consultation in clinical neurology which employs human-like reasoning in diagnosis, covers a broad diagnostic domain, has the capacity to explain strategies and conclusions and is readily expandable. Neurologist employs the strategy of first localizing a neurologic disease, then uses these data as well as mode of disease onset, rapidly to focus on a limited number of diagnostic possibilities which are then sequentially investigated. This paper presents the results of the preliminary assessment of Neurologist's diagnostic accuracy. For a set of 30 test cases, the system's leading diagnosis was correct in 77%, and separation of correct and incorrect diagnostic hypotheses was excellent. Further evaluations of the system are in progress.

House Officer Information System--HOIS

The design and operation of a clinical information system called HOIS (House Officer Information System) is described. Information about the management of 78 common acute medical problems is available to the user from this system. An analysis is presented of the medical problems found in 631 medical in-patients. It is reported that 98% of patients had at least one of the 78 problems found in HOIS's data base, and that 82% of all problems in these patients were found in this database.

Mathematical and computer assisted procedures in clinical decision making

Numerous mathematical and computer assisted procedures have been developed and tested as aids in clinical decision making. With pressures to curtail unnecessary utilization of diagnostic tests, these models may play an increasingly important role. In practice, additional data may benefit the construction of mathematical models but may not necessarily benefit clinicians. With improvements in computer technology, laboratory medicine is in a strategic position to influence the direction of diagnostic testing and clinical decisions in a more cost effective manner.

Computer-aided medical diagnosis: literature review

The difficulty of the medical diagnostic task and the advantages of the computer as an aid in this task are discussed. The general strategy and structure of any computer-aided system is presented, and the relationship of diagnostic accuracy to key variables involved in the development, test and use of a computer-aided diagnostic system is examined. These variables include: the computer algorithm, the source of the information used to develop the data base, the number and type of diseases under investigation, the number and type of indicants used, the source of the test sample, and the source of the validated diagnosis. A table of 58 empirically tested computer-aided medical diagnostic systems is presented; each system is summarised in relation to the variables mentioned above and diagnostic accuracy.

Optimizing the clinical value of hepatic scintiphotography

Hepatic scintiphotography serves as a highly useful anatomic liver function test. When the descriptive features of the liver scan are carefully related to underlying pathology, the resulting data base permits a detection sensitivity of 90% for metastatic cancer and for the diffuse parenchymal diseases. The consequences of false positive (FP) interpretations for patient care must be minimized by attention to perceptual and technical details, by incorporating other tests into the evaluative algorithm, and by knowledge of the distribution of diseases in the population examined. The liver scan would be more beneficially applied to patient management if referring physicians were made aware of its capabilities. They, in turn, must develop more explicit strategies for disease management in order that scintiphotography be effectively applied in clinical decision making.

Some observations on fuzzy diagnosis and medical computing

Wechsler has advocated the use of fuzzy logic for medical diagnosis, but it is pointed out that simpler systems have proved more successful than is apparent from his paper. The proposal raises a broad range of issues concerning the role of sophisticated information-processing techniques in the development of medical computing. These issues are discussed and it is suggested that a more evolutionary view of this development would be of greater benefit in the long run.

A disease-conscious method for sequential diagnosis by use of disease probabilities without assumption of symptom independence

A method for automatic diagnosis of disease is formulated and applied to a data base of several hundred gastroenterological patients who were each known to have one of six diseases. Application of the method requires no assumptions regarding statistical independence of symptoms. Each disease is associated with its own disease-symptom function, and any order of dependence between the symptoms and each disease may be allowed for. A patient's symptoms are used to determine the value of any specified disease-symptom function. This value is then used to determine the probability that the patient has the corresponding disease. The method is applied to a sequential diagnosis of patients not contained in the initial data base. Additional symptoms are chosen according to their diagnostic value. The entire model is disease-conscious in that disease-symptom functions, disease probabilities and diagnostic values need be evaluated only for those diseases that are considered relevant to the diagnosis.

Screening--an efficient tool in outpatient endocrine care. Studies of the diagnostic process

A screening procedure based on physical examination alone was used in the outpatient unit of a department of endocrinology and metabolism. The effectiveness, efficiency, and volume of this diagnostic process were evaluated by a study of the case records and a mail questionnaire. In 1968, 296 subjects were examined by this screening procedure; 103 of these could immediately be referred back. The percentage of confident endocrine diagnoses increased from 31 to 39%. The corresponding figure for the group of subjects immediately referred back was 100%. A considerable change in the diagnoses was also recorded. The use of the screening procedure saved approximately ately +30,000. It is concluded that diagnostic screening by an experienced endocrinologist is inexpensive, efficient and safe. The use of the screening procedure made it possible to offer highly specialized medical treatment to more people.

Diagnostic process and rheumatic diseases

The diagnostic process can be described as a continuous interaction between doctor and layman, whether neither computers nor tests represent final solutions. In this process, exchange and comparison of disease concepts and their components take place. The situation is analysed from the viewpoints of clinical work and epidemiology in the case of the rheumatic diseases, especially rheumatoid arthritis. The layman's part in the diagnostic process is considerable, both in terms of quantity and quality. It is concluded that our knowledge is strikingly limited and that future study should be devoted to doctor and patient as a unit when they together create diagnosis. Furthermore, international comparability of results and measurement of health levels require an expansion of our knowlege of the diagnostic process.