AI systems in clinical practice

Decision support systems
PAIRS
Physician assistant Artificial Intelligence System
Diagnostic decision support system

developed by clinical domains keywords
Dr. AM Mohan Rao, Logic Medical Systems, Hyderabad, India Multiple. Diagnosis of difficult cases. General practice Decision support systems, Bayesian probabilistic belief networks, variational methods for efficient inference in large-scale probabilistic models
location commissioned status
In use in Dr. AM Mohan Rao's clinic, Hyderabad, India 2001 In clinical use / under evaluation
description

PAIRS (Physician assistant Artificial Intelligence System) is designed to help doctors diagnose difficult cases. It is currently (October 2003) being tested in clinical practice in Hyderabad and its commercial launch is planned for January 2004.

Click on image for full size screeen
PAIRS screen
PAIRS screen

The PAIRS system is based on the variational methods for efficient inference in large-scale probabilistic models developed by Jaakkola and Jordan (1999). PAIRS works on a large database of over 30,000 disease-features and 620 internal medicine diseases. Each feature is quantified on the basis of its patho-physiology, disease incidence and possibility of it being caused by a disease not in the list. PAIRS includes a 7282-disease list, over 10,000-feature list and 415,000 feature-feature links. The clinical knowledge database has been built up from standard texts and peer-reviewed journals over the last eight years. One can access either disease-feature, feature-disease and feature-feature links or run a diagnosis from the front end.

A notable feature of the system is that it can provide a diagnosis for patient data including over 50 features. The accuracy of PAIRS is checked by patient data of each of 340 cases from Massachusetts General Hospital (published by New England Journal of Medicine). Some of the test cases have a large number of features, which is a limiting factor for AI systems based on Bayesian probabilistic belief networks. However, PAIRS can cope with such large patient data by partially doing an exact inference involving upper and lower bounds of quantifications of feature given disease. This process is known as variational transformation and involves finding conditional probabilities (exact and transformed) of features given disease. Only those (about 50% of positive findings) that have minimal difference between exact and transformed conditional probabilities are transformed and rest are treated exactly. The system calculates probabilities for each of diseases and gives a differential diagnosis, which eventually helps physician find an accurate diagnosis. A custom dictionary and a natural language processing (NLP) interface make entry of the patient data fully compatible with PAIRS database. This interface builds a word pool from translations and break-ups of Greek and Latin medical terms and finds their equivalent synonyms and antonyms before searching the database. Thus, based on the patient data a comprehensive case data is built on which the AI system works.

references
Rao, MM., Quantification of propensity for differential diagnosis by denotational semantics: logic formalisms for probabilistic model. 1999(Apr).P89-98. Paper presented at the IPE (Institute of Public Enterprises, Hyderabad) National conference on Medical Informatics.

[Paper]

" We report the design of logic formalisms for simplification of computations involved in belief network representation of Quick Medical Reference-Decision Theoretic (QMR-DT) diagnostic support tool. We quantified the features using denotational semantics. "Propensity for differential diagnosis' is defined as the ratio of sums of quantities of features shared by a group of diseases and their incidence in a given disease. We suggest that Ramsey's theorem is a definition for differential diagnosis. Simple Bayesian probabilistic method is used for testing the logic formalisms in our knowledge base. Several published reports on missed diagnosis give posterior probabilities of presumptive and missed diagnoses. Results derived from logic based and simple probabilistic model suggest that 'propensity for differential diagnosis ' concept can reduce the complexity involved in the computations of probabilistic belief networks of diagnostic systems. "
Rao, MM., Simulation of a virtual physician for diagnosis: Novel applications of genetic code for interlinks in data bank, Internet and citizen card technologies. 2000(Feb). P107- 111. Paper presented at the IPE (Institute of Public Enterprises, Hyderabad) National conferences on Medical Informatics. Summary: "We designed a computer based medical diagnostic system for logic formalisms of the Bayesian probabilistic belief networks. Data bank of about 35,000 disease-features is developed from text sources using MS Access. Its characteristics include quantification of propensity for differential diagnosis (ND) whose theoretical aspects are presented at the 1999 IPE Medical informatics conference. The program also includes a resident form, which is similar to Office assistant and gives the information on many diseases, features and their differential diagnoses using the hyperlink word documents. We developed feature to disease, disease to feature and feature to feature links. The pivot form gives the diagnosis for any patient data after calculating their probabilities. For faster computations and inter-linking the medical data we developed mathematical expressions for Genetic code, which yields 262,144 unique nine alphabet sequences formed by A,G,C and T alphabets. Similarly, eighteen alphabet sequences can give 68.7 billion unique sequences. We propose that a 32- bit processor and sufficient memory only is required in developing instantaneously connecting Internet and Citizen card technologies. We can reduce the volume of human genome sequence to one third since 9 alphabets can be replaced by three alphabets and a number. "
Rao, MM., Cyber Doc: Windows for medical diagnosis and data. 2000(Nov). Papers presented at the IPE National conferences on Medical Informatics.

[Abstract]  [Paper]

" Cyber doc is the first ever computer program that can diagnose a disease for the given features. It can give a range of disease possibilities for any given features. It also has an Office assistant that helps in giving any medical information. Doctors, medical students, patients and patients relatives can get instantaneous online information like diagnosis and diagnosis of difficult cases for any clinical data. This information is useful for early diagnosis, minimizing the investigations or as a second opinion. This program is also useful to healthcare personal, nurses, nursing students or health insurance people. "
Jaakkola, T.S. and Jordan, M.I. (1999). Variational Probabilistic Inference and the QMR-DT Network. J. Artificial Intelligence Research,10, 291-322.

[Paper (JAIR)]

" We describe a variational approximation method for efficient inference in large-scale probabilistic models. Variational methods are deterministic procedures that provide approximations to marginal and conditional probabilities of interest. They provide alternatives to approximate inference methods based on stochastic sampling or search. We describe a variational approach to the problem of diagnostic inference in the `Quick Medical Reference' (QMR) network. The QMR network is a large-scale probabilistic graphical model built on statistical and expert knowledge. Exact probabilistic inference is infeasible in this model for all but a small set of cases. We evaluate our variational inference algorithm on a large set of diagnostic test cases, comparing the algorithm to a state-of-the-art stochastic sampling method. "

Heckerman, D., Schacter, R., (1995). Decision-theoretic foundations for causal reasoning. J. Artificial Intelligence Research, 3, 405 430.

[Paper - Microsoft]

" We present a definition of cause and effect in terms of decision-theoretic primitives and thereby provide a principled foundation for causal reasoning. Our definition departs from the traditional view of causation in that causal assertions may vary with the set of decisions available. We argue that this approach provides added clarity to the notion of acyclic graphs. We describe a special class of influence diagrams, those in canonical form, and show its relationship to Pearl's representation of cause and effect. Finally, we show how canonical form facilitates counterfactual reasoning. "
Shwe MA.,Middleton B., Heckerman DE., Henrion M., Horvitz EJ., Lehman HP., Cooper GF. Probabilistic diagnosis using a reformulation of the INTERNIST-1/QMR knowledge base, part I. The probabilistic model and inference algorithms. Methods Inform Med 1991: 30:241-50.

[PubMed]  []

" In Part I of this two-part series, we report the design of a probabilistic reformulation of the Quick Medical Reference (QMR) diagnostic decision-support tool. We describe a two-level multiply connected belief-network representation of the QMR knowledge base of internal medicine. In the belief-network representation of the QMR knowledge base, we use probabilities derived from the QMR disease profiles, from QMR imports of findings, and from National Center for Health Statistics hospital-discharge statistics. We use a stochastic simulation algorithm for inference on the belief network. This algorithm computes estimates of the posterior marginal probabilities of diseases given a set of findings. In Part II of the series, we compare the performance of QMR to that of our probabilistic system on cases abstracted from continuing medical education materials from Scientific American Medicine. In addition, we analyze empirically several components of the probabilistic model and simulation algorithm. "
Middleton B., Shwe M., Heckerman D., Henrion M., Horvitz E., Lehman H., Cooper G. Probabilistic diagnosis using a reformulation of the INTERNIST-1/QMR knowledge base, part II: evaluation of diagnostic performance. Methods Inform Med 1991:30:256-67.

[PubMed]  [SMI]

" We have developed a probabilistic reformulation of the Quick Medical Reference (QMR) system. In Part I of this two-part series, we described a two-level, multiply connected belief-network representation of the QMR knowledge base and a simulation algorithm to perform probabilistic inference on the reformulated knowledge base. In Part II of this series, we report on an evaluation of the probabilistic QMR, in which we compare the performance of QMR to that of our probabilistic system on cases abstracted from continuing medical education materials from Scientific American Medicine. In addition, we analyze empirically several components of the probabilistic model and simulation algorithm. "

contact links

Dr. AM Mohan Rao, Logic Medical Systems, 68 Santosh Nagar, Mehidipatnam, Hyderabad-500 028, India.

amohanraoatrediffmail.com
 bullet  PAIRS
acknowledgements
Dr. AM Mohan Rao

Entry on archive: 23 October 2003
Last main update: 26 October 2003
Search this site
 

 

Privacy policy User agreement Copyright Feedback

Last modified:
© Copyright OpenClinical 2002-2011