Fred Wiener

Computer simulation of the diagnostic process in medicine.

Abstract A system based on the logical relations between a disease and its associated clinical findings is presented. For each disease, the associated clinical findings are separated into smaller units representing successive degrees of diagnostic certainty and consistent with the sequence in which the findings become known. Threshhold logic is used to express the Boolean combinations of findings sufficient for confirming a given diagnostic stage. The diagnostic process in a given area of medicine is formulated in a modular fashion, where each module consists of the conclusion (diagnostic stage or suggested clinical tests) to be reached and the factors (clinical findings or previous conclusions) upon which the conclusion depends. By dividing the often complex and seemingly instinctive inferences used in clinical practice into a series of logical steps which can be formulated for the computer, it is possible to simulate the physicians thought processes in reaching a diagnosis. The system also includes the computer programs to convert the medical logic modules into the corresponding files in computer memory, to allow entry of patient data and to analyze this data and report the appropriate conclusions.

Computerized medical reasoning in diagnosis and treatment of acid-base disorders

A system is described for aiding the clinician in the management of acid-base disorders. The medical knowledge required for interpretation of blood gas measurements, etiologic diagnosis, and treatment selection for acid-base disorders is structured into decision pathways consisting of a series of inferences. Each inference is defined by a medical logic module which specifies the different combinations of criteria, patient data and/or previously confirmed inferences, sufficient for confirming or rejecting the inference. A method is provided for converting numerical observations to the appropriate logical statement used in the modules. Patient data are compared to the medical logic and a status report lists the input data, acid-base diagnosis, and the suggested therapy. After initial testing on patient data, the medical logic was updated to express the medical policy of our clinical specialists. The system was applied to 54 patients and the systems conclusions were in full agreement with our staff in 93% of the cases, and in partial agreement in the other cases. The modular structure of the systems medical knowledge allows full expression of all the nuances of medical policy in our unit and facilitates updating to encompass the latest developments in acid-base management. The system can be integrated readily into existing computerized patient monitoring systems.

SMR (simulating medical reasoning): an expert shell for non-AI experts

SMR is an expert system shell designed to put the tools for knowledge acquisition directly into the hands of the domain expert. Since the knowledge base is represented as free text within a simplified syntactic structure, it is intelligible to anyone familiar with medical terminology. The knowledge base includes the rules for inference making as well as data groupings and protocols to facilitate case recording. In this paper, SMR is presented from the experts point of view, describing the rule syntax and procedures for formulating the required diagnostic or therapeutic knowledge in the chosen domain. Similarly, the end-users application of the system to patient data and the provisions for exploring and explaining the systems conclusions and reasoning processes are detailed. Avoiding tedious and often inane dialog, the user enters all that is known about the patient and receives a report of the systems conclusions and recommendations followed by a list of observations to be made in patient follow-up. An expert system for evaluating the diabetic patient is used to illustrate system operations.

Bone Browser a decision-aid for the radiological diagnosis of bone tumors

Bone Browser, a decision-aid for radiological diagnosis of bone tumors, was developed in cooperation with the Radiology Department of the Rambam Medical Center, Haifa. The system offers case specific advice from an expert system (ES), general information on the lesion area and allows for recording and retrieving cases. The ES utilizes both rule-based and probabilistic inferencing methodologies to arrive at a differential diagnosis (DD). The knowledge base was validated on 105 cases with known outcome. Clinical evaluation consisted of 59 new cases whose final diagnosis was not known to the evaluators. The correct diagnosis was included in the systems DD in 85% of the cases, which is comparable to the diagnostic accuracy of senior radiologists (88%). The system proved to be helpful to the expert, diagnosing cases missed by the radiologists and suggesting additional diagnoses not listed by the radiologists, raising their diagnostic capability to 91%.

Computerized classification of congenital malformations using a modified Bayesian approach

The diagnostic classification of children with dysmorphic features involves over 200 syndromes and 232 findings, with an average of about 15 findings per syndrome. A knowledge base expressed in terms of Boolean combinations of findings is impractical. The normal Bayesian method requires a very large incidence matrix with the vast majority of cells being zero. A modified Bayesian method is proposed in which each syndrome is described in terms of its associated findings, whose incidence P (S/D) are designated as essential (0.90), prevalent (0.90), occasional (0.70) or rare (0.15), whilst P(S/-D) ranged from (0.08) to (0.10). The Bayesian calculation determines the probability of the presence P(D/S) or the absence P(-D/S) of each syndrome. The differential diagnosis consisted of all syndromes whose presence has a probability greater than 0.85. One hundred and thirty-one cases from the Hanna Khoushi Developmental Pediatrics Center at Haifas Rothschild Hospital were considered. Of the 42 cases for which the centers specialists reached a diagnosis, the system listed the correct diagnosis for 91%. The system reached a diagnosis in about half of the remaining 89 cases. The medical literature is arranged by syndrome whilst the computer allows a case by case approach, thereby avoiding the need for the physician to consider each syndrome to see if it fits his case. This study shows that our modified Bayesian analysis is a valid method for shortening the physicians search in an area of great diagnostic complexity.

Mathematical model to study fluid and protein transfer in pulmonary edema in man

We developed a mathematical model that integrates the various processes that affect fluid and protein transport in the lung. The model is based on experimental data and current concepts of fluid and protein transfer in the lung, with particular attention to experimental studies of lymph flow and protein content. The model is exercised with data obtained from 5 patients with fulminant pulmonary edema who are studied prospectively. Tracheobronchial fluid (TBF) was sampled sequentially during the course of edema. In addition, radioiodinated human serum albumin (RIHSA) was injected iv and its appearance was measured in TBF as well as in plasma (P). The oncotic pressure of P and TBF, together with the measured pulmonary vascular hydrostatic pressure, and the appearance of RIHSA in TBF were used to exercise the model. Water flux across the microvascular wall was calculated using the Starling equation, and the equation for protein transport utilized terms for convection and diffusion. Transport coefficients were estimated which produced the closest fit between the mathematically calculated and measured values for TBF protein concentration and the rate of RIHSA appearance in TBF for each patient.For patients with permeability edema, our mathematical model estimated that bulk flow coefficients, Kf for water flux and (1 – σ) for convective flow of protein increased while the estimate of the coefficient for protein diffusion ω was increased to a lesser extent. In hemodynamic edema, the model predicted Kf and ω to be somewhat elevated, but because the estimated σ was unchanged, there was no massive loss of protein from the vascular space. The measured values and the values calculated by the model for TBF/P protein ratio averaged 0.66 for hemodynamic edema, and 0.94 for permeability edema. The model predicted that a maximal increase in lymph flow, consistent with experimentally measured values, would substantially reduce extravascular volume in hemodynamic edema within 24 h, but would have little effect in permeability edema. One explanation for the observed greater concentration of tagged protein in the TBF than in P in 1 patient is provided by the model. The time after RIHSA injection when this crossover is predicted by the model averaged 20 h in hemodynamic edema, but only 4 h in permeability edema. The duration of this time interval may be a useful variable to distinguish the 2 types of edema.We recognize the shortcomings of a model based on disparate experimental data. However, we believe that the model provides a method to help improve our understanding of edemagenesis. The model gives the clinical investigator a prediction of what may follow the manipulation of one or more of the multiple variables that influence edemagenesis. As additional clinical and experimental knowledge of edemagenesis accumulates, the model can be refined.

Computer systems for facilitating management of the critically ill

The Shock Research Unit has applied computer technology to the care of the critically ill and injured patient since 1961. The requirements for patient monitoring were initially explored with a process control computer (IBM) 1710). In the current system, a Xerox Sigma-5 computer is utilized for monitoring EKG, hemodynamic, respiratory, and biochemical signals. Electronic preprocessing increases the efficiency and speed of data acquisition and signal analysis. Provisions are made for recording narrative data as part of a commitment to evolve an automated patient record. Bedside displays include both tabular and graphic summaries of patient status and trends. A computer accessible archive of patient files is maintained. Clinical operation of the system has been facilitated by automation of afferent and efferent functions including flushing of catheters, servo-calibration of pressure measuring systems, automated urine collection and disposal, and computer controlled infusion of fluids and medications. We anticipate the continuing development of automated afferent and efferent components for feedback control of ventilators by automated sampling and measurement of arterial blood gases and infusion of fluids and medications in response to changes in monitored hemodynamic variables. Such automation, together with medical intelligence for priority alarms and interpretive displays, hold promise of increasingly potent and cost-effective systems to facilitate and improve care of the critically ill or injured patient.

Computer simulation of medical reasoning in the assessment of kidney disease

A system for the simulation of clinical reasoning was applied to the evaluation of patient data in nephrological diseases. The system allows the physician to computerise the medical logic in his area of specialisation, to prepare suitable protocols for recording clinical observations, to analyse the patient data for assessing his present clinical status and to plan appropriate diagnostic and therapeutic interventions. The medical logic is expressed in modular form as a series of inferences to be confirmed or rejected on the basis of Boolean combinations of clinical findings and previous inferences. The acceptable Boolean combinations are formulated in terms of threshold logic. The computer programs produce alphabetical lists of the clinical findings appearing in the medical logic, which are organised by the physician into data acquisition protocols appropriate to each phase of clinical activity. The system was run on a series of 22 patients hospitalised in the Department of Nephrology. The patient status reports produced by the computer were in substantial agreement with the assessment of a senior nephrologist. The protocols assured a more complete and accurate recording of patient data than that appearing in the routine patient chart. The program was of value in standardising patient examinations and in the training of new medical staff.

A system for the computer simulation of clinical reasoning.

Abstract A system is presented for computer simulation of the logical processes used by the physician in clinical medicine. The medical logic in a given area is formulated in modular fashion, where each module states the inference to be drawn and the clinical data on which the inference is based. The complex reasoning in diagnosis or therapy is reduced to a series of logical inferences each of which depends on a small number of clinical findings. Since all findings will not necessarily be present in each patient, the Boolean subsets of the findings sufficient for confirming the inference are expressed by threshold logic. A numerical weight is assigned to each finding such that the sufficient subsets will have a total score exceeding a preset threshold. The system programs are: EDIT, to convert the physician formulated medical logic modules into appropriate computer files; LIST, to aid in correcting and standardizing the medical terminology; QUERY, to allow on-line entry of patient data; and TEST, to analyze the patient data and report the patients status.

Computerized medical decisions in evaluating the diabetes patient

A system for computer simulation of medical reasoning is described which has been applied to classifying diabetic patients and assessing the complications due to disease. Medical knowledge is formulated into modular, interrelated inference pathways. Each module states the inference and the Boolean combinations of criteria sufficient for confirming or rejecting it. Patient data is compared to the medical logic and a status report is produced which presents the clinical findings, confirmed inferences, the diabetes evaluation score for each body system and request for additional data at the follow-up visit. After initial testing the medical logic was easily updated to fully express medical policy in our clinic. The system provides an objective and consistent method for evaluating the diabetic patient.