From widman@CAMIS.Stanford.EDU Fri Mar 12 21:19:20 1993 Return-Path: Received: from medmail.Stanford.EDU by HPP.Stanford.EDU (4.1/inc-1.0) id AA08550; Fri, 12 Mar 93 21:19:20 PST Received: from CAMIS.Stanford.EDU by medmail.Stanford.EDU (4.1/inc-1.0) id AA23950; Fri, 12 Mar 93 21:18:51 PST Received: by CAMIS.Stanford.EDU (4.1/inc-1.0) id AA16475; Fri, 12 Mar 93 21:19:16 PST Sender: Lawrence Widman Date: Fri, 12 Mar 1993 21:19:15 PST From: Lawrence E. Widman Reply-To: widman@sumex-aim.stanford.edu To: ai-medicine@med.stanford.edu Subject: SUMMARY: Cardiology applications of ES/neural nets Message-Id: ORIGINAL MESSAGE: Sender: Lawrence Widman Date: Wed, 10 Feb 1993 16:40:24 PST From: Lawrence E. Widman Reply-To: widman@sumex-aim.stanford.edu To: ai-medicine@med.stanford.edu Subject: Cardiology applications of ES/neural nets I will be giving a summary-type talk on expert systems and neural networks at the American College of Cardiology meeting in March. This is one of the two international clinical cardiology meetings held in the US each year. The venue will be a symposium on computers in cardiology. Please let me know of any practical, working applications of expert systems and/or neural network technologies in clinical cardiology and related areas. Also interesting would be research projects that have innovative clinical implications. I will summarize to the net. Many thanks. -- Larry Widman --------------------------------------------------------------------- RESPONSES: --------------------------------------------------------------------- 11-Feb-93 16:16:36-GMT,1292;000000000000 Date: Thu, 11 Feb 93 10:18:00 CST From: kford@trivia.coginst.uwf.edu Subject: Re: expertsystem in cardiology >Ken, > I would be very interested in seeing the report on NUCES. Could you send >it by email? I can read LaTeX, WordPerfect, ASCII, and any other >reasonable format. Thanks. -- Larry Larry, The tech report is on a Mac in a word processor called NISUS. It contains some screen shots that are each over a megabyte of memory. Thus, perhaps I should federal express it to you if time is a consideration. What address should I use? Cheers, Ken ________________ Ken Ford (904) 474-2551 (Office) Institute for Human & Machine Cognition (904) 474-3023 (FAX) Division of Computer Science kford@ai.uwf.edu (internet) University of West Florida 11000 University Parkway Pensacola, FL 32514 --------------------------------------------------------------------- 11-Feb-93 17:57:57-GMT,4845;000000000000 Date: Thu, 11 Feb 93 12:01:04 GMT From: jeremy@bison.lif.icnet.uk (Jeremy Wyatt) Subject: Re: Cardiology applications of ES/neural nets Larry, >Please let me know of any practical, working applications >of expert systems and/or neural network technologies in >clinical cardiology and related areas. Also interesting would >be research projects that have innovative clinical implications. I attach a description of the ACORN system; I'm no longer in touch with the project, but ACORN was certainly in routine use at Westminster during the period 1987-90. Good luck with your lecture. Best wishes, Jeremy Wyatt MD MRCP(UK) DM(Oxon), Biomedical Informatics Unit, Imperial Cancer Research Fund, PO Box 123, Lincoln's Inn Fields, London WC2A 3PX, UK. Tel: +44 (0)71-269-3637 Fax: +44 (0)71-430-1787 -------------------- System name: ACORN (Admit to the Ccu OR Not) System: Hybrid rule-based & Bayesian system for advising on management of chest pain patients in the emergency room Location: Accident & Emergency Department, Westminster Hospital, London Contact: Jeremy Wyatt, wyatt@sumex-aim.stanford.edu Date commissioned: 1987 Output: Text report on the patient's diagnosis and need for admission to CCU Decription: Medical audits determined that 38% of patients attending a teaching hospital emergency room with acute ischeamic heart disease were sent home in error, and that the median time till CCU admission for the remainder was 115 minutes. We built ACORN, a hybrid backward-chaining rule-based and Bayesian system, to assist senior nurses and casualty officers in the management of these patients. In a blinded randomised controlled trial of 153 patients in 1987 the false negative rate (patients with acute ischaemic heart disease accidentally sent home) in both control & ACORN cases fell to 20%. ACORN was used in all study cases and in at least 5 of the control cases; the change in false negative rate in the control group may thus have been due to use of ACORN, simple carryover or to the Hawthorne effect. The trial also identified significant problems with ACORN, and it was subsequently revised. The revised version appeared to reduce the median time to CCU admission by 20 minutes, though this was an uncontrolled study. In 1990, nearly three years after the field trial, the users claimed to use ACORN in 77% of eligible cases, but unequivocal evidence of use was present in the patient records in only 23% of eligible cases. There were approx 15 eligible cases per week (750 per year), so this scales up to between 175 and 580 uses of ACORN per annum in 1990. Refs: - Wyatt J. "The evaluation of clinical decision aids: a discussion of methodology used in the ACORN project", Lecture Notes in Medical Informatics 1987; 33: 15-24. - Wyatt J, Dillistone J, Emerson PA (1988). Thrombolysis in acute MI: why the delay ? Brit. Heart J. 1988; 59: 618. (abstract) - Wyatt J (1989). Lessons learned from the field trial of ACORN, an expert system to advise on chest pain. In: Barber B, Cao D, Qin D, eds. Proc. Sixth World Conference on Medical Informatics, Singapore. Amsterdam: North Holland 1989: 111-115 - Wyatt J, Emerson P : "A pragmatic approach to knowledge engineering, with examples of use in a difficult domain." Chapter 4 in "Expert Systems: Human Issues", eds. Berry D, Hart A. London: Chapman & Hall / MIT Press, 1990. - Emerson PA, Russell NR, Wyatt JC et al (1989). An audit of the management of patients attending an accident and emergency department with chest pain. Quart J Med 1989; 70: 213-220 - Hart A, Wyatt J (1989). Connectionist models in medicine: an investigation of their potential. Proc. Second European Conference on Artificial Intelligence in Medicine; Hunter J, Cookson J, Wyatt J (eds). Heidelberg 1989: Springer Verlag - Wyatt J. "A method for developing medical decision-aids applied to ACORN, a chest pain advisor". DM Thesis, Oxford University, 1991 - Wyatt J, Spiegelhalter D. Field trials of medical decision-aids: potential problems and solutions. In Clayton P (ed). Proc. 15th Symposium on Computer Applications in Medical Care, Washington 1991. New York: McGraw Hill Inc 1991: 3-7 - Wyatt J. "Computer-based knowledge systems". The Lancet 1991; 338: 1431-1436 - Wyatt J, Spiegelhalter D. The evaluation of medical expert systems. In: Evans D, Patel V (eds), Advanced models of cognition for Medical Training and Practice (NATO ASI series F97). Heidelberg: Springer Verlag 1992: 101-120 --------------------------------------------------------------------- 11-Feb-93 19:59:48-GMT,1645;000000000000 Date: Thu, 11 Feb 93 14:09:34 -0600 From: blaz@cs.uh.edu (Blaz Zupan) Subject: RE: Cardiology applications of ES/neural nets Hello! There was a nice application build in Ljubljana, Slovenia by Prof. Ivan Bratko group. The application name is KARDIO and it was an expert system based on a qualitative model. The language used was Prolog. I don't know whether the application is now still in use, but I am sure that University Clinical Center in Ljubljana was using it some time ago. There was a book published about this application (I may be wrong, but I think that the publisher was Addison-Wesley. The title of a book is KARDIO and the author is I. Bratko). Let me know if you are interested about the exact info about this book or the application. Regards, Blaz Zupan (blaz@cs.uh.edu) [Ed: The book is: Book review of: Ivan Bratko, Igor Mozetic, and Nada Lavrac. "KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems." MIT Press, 1989. See also a review in Artificial Intelligence in Medicine. 1991; 3:359--360. ] --------------------------------------------------------------------- 11-Feb-93 21:58:29-GMT,1223;000000000000 Date: Thu, 11 Feb 1993 13:57 PDT From: "TIM DENTON, CVS, EXT. 3851" Subject: ai-medicine@med.stanford.edu mailing list I just received a forwarded message from someone regarding your talk at the Anaheim meetings. We have a few projects going on comparing the predictive ability of logistic regression, bayesian belief networks and neural nets. It sounds like your talk would be quite appropriate. Hope to see you there. Tim Cardiothoracic Surgery Cedars-Sinai Medical Center, LA --------------------------------------------------------------------- 11-Feb-93 22:30:11-GMT,1082;000000000000 Date: Thu, 11 Feb 93 10:12:25 GMT From: Kathleen King Organisation: Dept. of Artificial Intelligence, Univ. of Edinburgh. Subject: Re: Cardiology applications of ES/neural nets You might contact Rob Harrison CO1RFH%primea.sheffield.ac.uk about his neural net cardiology thing (working, in use). kathleen --------------------------------------------------------------------- 12-Feb-93 15:18:45-GMT,1164;000000000000 From: David Spiegelhalter Date: Fri, 12 Feb 93 09:15:05 GMT Subject: ES in cardiology Dear Professor Widman, We have built a Bayesian network for diagnosis in congenital heart disease, in conjunction with Great Ormond Street Hospital departemnt of Paediatric Cardiology. In fact we are currently doing a comparison of expert-built network, data-built network, expert-built algorithm, data-built algorithm, logistic regression and neural networks! Some of these have been written up - I could send you some material if you let me have your full address. Best wishes David Spiegelhalter --------------------------------------------------------------------- 11-Feb-93 3:48:57-GMT,1989;000000000000 Date: Wed, 10 Feb 93 20:48:55 -0700 From: soller@asylum.cs.utah.edu (Jerome Soller) Subject: Re: Cardiology applications of ES/neural nets Dear Dr. Widman: Dr. Charles Rosenberg of the VA GRECC and the University of Utah Department of Psychology has done some work on Neural Network models using myocardial thallium scintigrams as the data. His work was done when he was in Israel before he joined us, and I have given him a copy of your request. His work will be published in a future edition of the Neural Computation Journal, (June, I think) and in the proceedings of the Neural Information Processing Systems Conference. His e-mail is crr@cogsci.psych.utah.edu . Dr. William Baxt of UCSD has developed another famous neural network, which he presented as the keynote talk of this year's NIPS conference. Many people from the neural network community consider him the main researcher in medical applications for the field. His phone number is 543-6463 (I don't know the area code, and he` does not know me personally). His work was published in the Annals of Internal Medicine and in Neural Computation. It is diagnosing myocardial infarction. The final one I may know about is Dr. Dean Sittig's (was from Yale, now from Vanderbilt University) work using a kalman filter and possibly a neural network for arrythmias or artifacting (I forgot it). For any of these, Dr. Rosenberg or I can find you the exact references. Sincerely, Jerome Soller University of Utah Department of Computer Science and VA GRECC --------------------------------------------------------------------- 11-Feb-93 4:20:26-GMT,3031;000000000000 From: sasi@saathi.ncst.ernet.in Subject: Re: Cardiology applications of ES/neural nets Date: Thu, 11 Feb 93 09:02:41 +0530 There is a cardiology expert system called Hruday developed using the expert system shell from NCST, Vidwan. Hruday's context is the following: Imagine you have been having chest pain for last few days. May be you notice you also feel tired very often. You start wondering if you have any heart trouble on the way. But you do not want to go to a cardiologist yet and blow up a heavy sum of money. Then you think of your friend who has just finished his studies in cardiology. You walk upto him for a chat... What ensues could be a dialogue with Hruday. The system does not use results of medical tests, etc. Mostly questions you can answer at any time anywhere such as characteristics of the pain, historical details, living style etc are used to see if you have something serious to worry about. The system has not been put to use in any medical context. But we have had medical doctors , informally, taking runs with the system and approving the results. We feel, minimally, it can have big educational value in medical colleges, etc. It could also be useful for households which have access to PCs to spread awareness of cardiac factors. The person primarily involved with the development is Dr S Ramani, Director, NCST. His e-mail address is 'ramani@saathi.ncst.ernet.in.'. There is a short paper on the system published in a local medical conference. We can send you a copy, if you are interested. You can also contact Dr Ramani at the above address. I would be very much interested in the summary. Please do post to the news group about your experience as well. - sasi +-------------------------------------------------------------------------+ |Sasikumar M, : E-mail- sasi@ncst.ernet.in| |National Centre for Software Technology,: Phone: 62 01 606 x 41 | |Gulmohar Cross Road 9, Juhu : ===============================| |Bombay - 400 049, INDIA : ===============================| +-------------------------------------------------------------------------+ --------------------------------------------------------------------- 11-Feb-93 4:34:21-GMT,1129;000000000000 Date: Wed, 10 Feb 93 22:35:43 CST From: kford@trivia.coginst.uwf.edu (Ken Ford) Subject: expertsystem in cardiology Hello, I am responding to your call for info on expert systems in cardiology. For that last several years we have been working on an expert system called NUCES (nuclear cardiology expert system). I could send you a brief tech. report if you are interested. Cheers, Ken Ford ________________ Ken Ford (904) 474-2551 (Office) Institute for Human & Machine Cognition (904) 474-3023 (FAX) Division of Computer Science kford@ai.uwf.edu (internet) University of West Florida 11000 University Parkway Pensacola, FL 32514 --------------------------------------------------------------------- 11-Feb-93 10:25:50-GMT,2869;000000000000 Date: Thu, 11 Feb 1993 11:24:18 +0100 From: Erich Prem Subject: Re: Cardiology applications of ES/neural nets Hi! The following are two references for two papers which describe research on developing a neural network that learns to diagnose coronary artery disease from thallium-201 scintigrams. It is novel in that the neural network training has been combined with ``symbolic'' knowledge as given by the medical expert. If you need the papers, let me know. Erich Prem. ----------- G.Dorffner, E.Prem, M.Mackinger, S.Kundrat, G.Porenta, H.Sochor "Experiences with Neural Networks as a Diagnostic Tool in Medical Image Processing." (TR-93-12 Austrian Research Institute for AI, Vienna) to appear in 37.Dt.Jahrestagung f"ur Med.\ Informatik, Biometrie und Epidemiologie (GMDS), Proceedings, 1993. E.Prem, M.Mackinger, G.Dorffner, G.Porenta, H.Sochor "Concept Support as a Method for Programming Neural Networks with Symbolic Knowledge." (TR-92-4 Austrian Research Institute for AI) to appear Proc.\ of the German Workshop on Artificial Intelligence (GWAI92), Springer Verlag, Berlin-Heidelberg, 1992. Neural networks are usually seen as obtaining all their knowledge through training on the basis of examples. In many AI applications appropriate for neural networks, however, symbolic knowledge does exist which describes a large number of cases relatively well, or at least contributes to partial solutions. From a practical point of view it appears to be a waste of resources to give up this knowledge altogether by training a network from scratch. This paper introduces a method for inserting symbolic knowledge into a neural network--called ``concept support.'' This method is non-intrusive in that it does not rely on immediately setting any internal variable, such as weights. Instead, knowledge is inserted through pre-training on concepts or rules believed to be essential for the task. Thus the knowledge actually accessible for the neural network remains distributed or --as it is called-- subsymbolic. Results from a test application are reported which show considerable improvements in generalization. ------------------------------------------ --------------------- Erich Prem Austrian Research Institute Tel. + 43 1 5336112 for Artificial Intelligence FAX + 43 1 5320652 Schotteng.3, A-1010 Vienna, AUSTRIA erich@ai.univie.ac.at --------------------------------------------------------------------- 11-Feb-93 14:44:08-GMT,9461;000000000000 Date: Thu, 11 Feb 1993 09:26:29 BSC (-0300 C) From: SABBATINI%ccvax.unicamp.br@Forsythe.Stanford.EDU Subject: Re: Cardiology applications of ES/neural nets Campinas, 11 Febr 1992 Dear Dr. Widman: Pleased to help you. I have been commissioned by CRC Reviews on Biomedical Engineering to write an extensive monograph on neural networks applications in Medicine, so I am collecting a lot of references on the subject. I am sending to you what I have got in Cardiology. I would appreciate very much if you send to me the references/abstracts you will receive from the net discussion. General reviews on neural networks in Medicine are: - Sabbatini, R.M.E. - Applications of connectionist systems in Biomedicine. Proceed. 7th World Congress on Medical Informatics (MEDINFO 92). International Federation of Medical Informatics. North Holland, Amsterdam, p. 418-426, 1992a. - Reggia, J.A. - Neural computation in Medicine. Artificial Intelligence in Medicine (in press), April 1993. Best regards Renato M.E. Sabbatini, PhD Head, Neural Networks Applications Group Director, Center for Biomedical Informatics Professor, Medical Informatics/School of Medicine State University of Campinas Campinas, Sao Paulo 13081-970, BRAZIL SABBATINI@CCVAX.UNICAMP.BR P.S.: I have not included references on clinical applications of cardiac imaging (such as SPECT, PET, etc.). Would you like to get them too ? I suggest also (if you didn't already) to peruse the proceedings of Computers in Cardiology conferences (IEEE Press). I gather that most of the recent work is there. ================================================================== CLINICAL CARDIOLOGY APPLICATIONS - Akay, M. - Noninvasive diagnosis of coronary artery disease using a neural network algorithm. Biol. Cybern., 67(4):361-7, 1992 - Avanzolini, G.; Barbini, P.; Gnudi, G. - Unsupervised learning and discriminant analysis applied to identification of high risk postoperative cardiac patients. Int. J. Bio-Med. Comput., 25(2- 3):207-221, 1990. - Akay, M.; Akay, Y.M.; Welkowitz, W. - Neural networks for the diagnosis of coronary artery disease. Proceed. Int. Joint Conf. Neural Networks, Baltimore, MD, June 1992. New York: IEEE Press, 1992. - Baxt, W.G. - Use of an artificial neural network for data analysis in clinical decision making. Neural Computation, 2: 480-9, 1990. - Baxt, W.G. - Use of an artificial neural network for the diagnosis of myocardial infarction. Ann.Intern. Med., 115: 843-848, 1991. - Browner, W.S. - Myocardial infarction prediction by artificial neural networks. Ann.Intern.Med., 116(8):701-2; discussion 702, 1992. - Cianflone, D.; Carlino, M.; Caradente, O.; Meloni, C.; Margonato, A.; Chierchia, S. - Neural network computing in Medicine: Realization of a succesful model for exercise stress evaluation. In: O'Moore, R.O. et al. (Ed.) - Medical Informatics Europe '90. Proceedings. Berlin: Springer-Verlag, p. 788, 1990. - Furlong, J.W.; Dupuy, M.E.; Heinsimer, J.A. - Neural network analysis of serial cardiac enzyme data. A clinical application of artificial machine intelligence. Am. J. Clin. Pathol., 96(1):134-41, 1991. - Harrison, R.F.; Marshall, S.J.; Kennedy, R.L. - A connectionist aid to the early diagnosis of myocardial infarction. Proceed. Third European Conf. Artificial Intelligence in Medicine, Maastricht, The Netherlands, June 1991a. - Harrison, R.F.; Marshall, S.J.; Kennedy, R.L. - The early diagnosis of heart attacks: a neurocomputational approach. Proceed. Internat. Joint Conf. Neural Networks, Vol.1: p. 1-5, 1991b. - Poli, R.; Cagnoni, S.; Livi, R.; Coppini, G.; Valli, G. - A neural network expert system for diagnosing and treating hypertension. Computer, 24(3): 64-71, 1991. ================================================================== ELECTROCARDIOLOGY APPLICATIONS - Bortolan, G.; Degani, R.; Willems, J.L. - Design of neural networks for classification of electrocardiography signals. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990. - Bortolan G., Degani R., Willems J. - ECG classification with neural networks and cluster analysis. In: Computers in Cardiology 1991, Murray, A.; Arzbaecher, R. (Ed.), IEEE Computer Society, Los Alamitos, CA, 177-180, 1991, 1991. - Bortolan, G.; Degani, R.; Willems, J.L. - Neural networks for ECG classification. In: Computers in Cardiology 1990, Murray A., Ripley K. L. (Eds.), IEEE Computer Society, Los Alamitos, CA, 269-272, 1991. - Carroll, T.O.; Ved, H.; Reilly, D. - A neural network for ECG analysis. IJCNN International Joint Conference on Neural Networks. New York, NY: IEEE, Vol. 2, p. 575, 1989. - Cheung, J.Y., Yeo, C.Y.S.; Hull Jr., S.S. - Pattern recognition using a bipolar associative memory. Proceedings 31st Midwest Symposium on Circuits and Systems, pp.165-168, August, 1988. - Cheung, J.Y.; Hull, Jr., S.S. - Detection of abnormal electrocardiograms using a neural network approach. Proceed. Ann. Int. Conf.IEEE Engineering in Medicine and Biology Society, vol 11, Part 6/6, pp. 201 5-2016, November, 1989. - Cheung, J.Y; Hull Jr., S.S.; Yeo, C.Y.S.; Kohli, P. - Recognition of abnormal EKG signals by a neural network approach. Int, J. Microcomp. Appl., 10(2):48-53, 1991. - Dassen, W.R.; Mulleneers. R.G.; Den Dulk. K.; Smeets, J.R.; Cruz, F.; Penn, O.C.; Wellens, H.J. - An artificial neural network to localize atrioventricular accessory pathways in patients suffering from the Wolff-Parkinson-White syndrome. PACE 13(12 Pt 2):1792-6, 1990. - Edenbrandt, L.; Devine, B.; Macfarlane, P.W. - Neural networks for classification of ECG ST-T segments. J. Electrocardiol., 25(3):167-73, 1992. - Evans, S.; Hastings, H.; Bodenheimer, M. - Self-training artificial neural networks can distinguish between beats of sinus and ventricular origin. Europ. J. Card.Pac. Electrophysiology, 2(2. Supl. 1A): A110, 1992. - Koska, M.; Vicenik, K.; Chudy, L. - Beat-to-beat detection of His-Purkinje system signals using neural networks. Proceed. Int. Symp. Neural Networks and Neural Computing, Prague, p. 205, Sept. 1990 - Farrugia, S.; Yee, H.; Nickolls, F. - Neural network classification of intracardiac electrograms. Europ. J. Card.Pac. Electrophysiology, 2(2. Supl. 1A): A109, 1992. - Iwata, A.; Nagasaka, Y.; Suzumura, N. - A digital Holter monitoring system with dual 3-layers neural networks. IJCNN International Joint Conference on Neural Networks. New York, NY: IEEE, Vol. 2, p. 69-74, 1989. - Iwata, A.; Nagasaka, Y.; Suzumura, N. - Data compression of the ECG using neural network for digital Holter monitor. IEEE Engineer. Med. Biol. Magaz., 9(3): 53-57, 1990. - Iwata, A.; Sugamoto, T.; Suzumura, N.; Moriguchi, Y. - A fast analyzing system for Holter recording using digital signal processing and neural networks. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990. - Koska, M.; Vicenik, K.; Chudy, L. - Beat-to-beat detection of His-Purkinje system signals using neural networks. Proceed. Int. Symp. Neural Networks and Neural Computing, Prague, p. 205, Sept. 1990 - Linnebank, A.C.; Sippens-Groenenweg, A.; Grimbergen, C.A. - Artificial neural networks applied in multiple lead electrocardiography. Rapid quantitative classification of ventricular tachycardia QRS integral patterns. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990 - Macerata, A.; Morabito, M.; Taddei, A.; Marchesi, C. - ANN approach for QRS morphological classification. In: Proceed. VI Mediterrean Conf. Med. Biol. Engineer. International Federation for Medical and Biological Engineering, Capri, Italia, p. 919-922, 1992. - Morabito, M.; Macerata, A.; Taddei, A.; Marchesi, C. - QRS morphological classification using artificial neural networks. In: Computers in Cardiology, IEEE Computer Society, p 181-184, 1991. Palmer, D. - Removing random noise from EKG signals. In: DARPA Neural Network Study, p. 588, 1988. - Macfarlane, P.W. - Recent developments in computer analysis of ECGs. Clin.Physiol., 12(3):313-7, 1992. - Rizos, G.; Anogianakis, G.; Apostolakis, M.; Nowack-Apostolakis, E. - Artificial neural networks (ANNs) that learn to recognize characteristic electrocardiographic (ECGs) patterns. In: Proceed. VI Mediterrean Conf. Med. Biol. Engineer. International Federation for Medical and Biological Engineering, Capri, Italia, p. 927-930, 1992 - Suzuki, Y.; Ono, K. - Personal computer system for ECG ST-segment recognition based on neural networks. Med. Biol. Eng. Comput., 30(1):2-8, 1992. - Tsai, Y.S.; Hung, B.N.; Tung, S.F. - An experiment on ECG classification using backpropagation neural network. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990 - Xue, Q.; Hen, Y.; Tompkins, W.J. - Training of ECG signals in neural network pattern recognition. Proceed. 12th. Ann. Conf. IEEE Engineer. Med. Biol. Soc., Philadelphia, PA, USA, 1990 --------------------------------------------------------------------- 12-Feb-93 18:44:02-GMT,913;000000000011 From: fmora@conicit.ve (Fernando Mora C. (USB)) Subject: ..on Card. ES and NC Date: Fri, 12 Feb 93 14:47:24 AST Dear Dr. Widman: How would you like this information, a list of papers or actual copies. Which areas are you more interested on?. Our group has developed some applications in the monitoring side. Particularly on ECG signal interpretation. Please Let me know. Fernando Mora, PhD --------------------------------------------------------------------- 12-Feb-93 22:13:49-GMT,3971;000000000000 From: Nancy Reed Date: Fri, 12 Feb 93 16:13:44 CST Subject: Re: Cardiology applications of ES/neural nets Dr. Widman, I don't know if this is the type of thing you are looking for or not. Below is a summary of my thesis research that will soon appear in the Focus column of the AI in Medicine newsletter (AAAI). Regards, -Nancy --------------------------------------------------------- Focus: Fallot Organization: University of Minnesota: Computer Science Department, Information and Decision Sciences Department, and the Pediatric Cardiology Department. Brief description of project: Nancy Reed's doctoral thesis research concerns the development of methods for diagnosing multiple interacting defects. Cases with multiple defects can be difficult to diagnose because the defects can interact, meaning that the observable cues are not the sum of the cues for the component defects. When defects interact, expected cues can be missing, cues may be altered, or new cues may appear. A complete quantitative model of the physiological states and processes can be used to simulate the effects of multiple defects, however many medical domains do not have an available model. This research examines a specific domain and develops a description and classification of the types of changes in cues produced by interactions between defects. We found that cues combine with one another in a small number of ways. Some ways involve simple general algorithms like set union, partial ordering, and arithmetic addition. Others involve more complex heuristics that are specialized for specific cues, tests, or defects. A computational diagnostic model is developed which includes methods that use the classification to reason about defect interactions. The model is tested by constructing a diagnostic program in the domain of pediatric cardiology, which diagnoses cases of multiple heart defects from their interacting cues. Application domain: Congenital heart defects, pediatric cardiology. Test results from auscultation, EKG, Xray, and the physical exam are used. Evaluation includes comparing the correct diagnosis with the original expert diagnosis, the diagnosis of an existing expert system (Galen), and the diagnosis of the new program (Fallot) on cases from hospital files. Specific attention is given to the following kinds of defects: septal defects, valvular stenosis, anomalous venous connections, and Tetralogy of Fallot. The finished program might be useful in an instructional setting by demonstrating a computational model of diagnosis. Contact addresses: Nancy Reed or Prof. Maria Gini, Computer Science Department, University of Minnesota, 4-192 EE/CSci Bldg, 200 Union Street S.E., Minneapolis, MN 55455 Phone: Reed (916) 757-2575, Gini (612) 625-5582 Email: reed@cs.umn.edu, gini@cs.umn.edu Prof. Paul Johnson, Information and Decision Sciences Department, University of Minnesota, 358 HHH, 271-19th Ave. S., Minneapolis, MN 55455 Phone: (612) 624-5570 Email: johnson@cs.umn.edu Related Publications: P. E. Johnson, A. Duran, F. Hassebrock, J. Moller, and M. Prietula, "Expertise and Error in Diagnostic Reasoning", Cognitive Science, Vol. 5, pp. 235-283, 1981. W. B. Thompson, P. E. Johnson, and J. B. Moen, "Recognition-based Diagnostic Reasoning", in Proc. IJCAI-83, pp. 236-238. P. E. Johnson, J. B. Moen, and W. B. Thompson, "Garden Path Errors in Diagnostic Reasoning", in Expert System Applications, Eds. L. Bolc and M.J. Coombs, pp. 395-427, Springer-Verlag, 1988. --------------------------------------------------------------------- 12-Feb-93 23:32:34-GMT,1143;000000000000 Date: Fri, 12 Feb 93 15:32:25 PST From: sirothe@srv.PacBell.COM (Sheldon Rothenberg) Subject: Re: Cardiology applications of ES/neural nets There was a paper presented at the Innovative Applications of AI '91 conference, whose proceedings sre published by AAAI Press, through MIT Press. The paper was The Thallium Disgnostic Workstation, developed by Rin Saunders. Other than that, I have heard of little systems, none of which I am sure has ever gone into production about assessing type A behavior, risk factors for chd etc... Shelley --------------------------------------------------------------------- 13-Feb-93 0:13:21-GMT,840;000000000000 Date: Fri, 12 Feb 93 15:59:25 From: jhunter@computing-science.aberdeen.ac.uk (Jim Hunter) Subject: Re: Cardiology applications of ES/neural nets Larry, I guess that you won't need reminding about our Ticker modelling. Best wishes, Jim Hunter [Ed: The citation is: Jim Hunter, Ian Kirby, and Nick Gotts. Using quantitative and qualitative constraints in models of cardiac electrophysiology. Artificial Intelligence in Medicine, 3:41--61, 1991. ] --------------------------------------------------------------------- 13-Feb-93 6:32:23-GMT,1285;000000000000 Date: 13 Feb 93 01:27:02 EST From: Chuck Kowalewski <72727.1770@CompuServe.COM> Subject: Re: Cardiology applications of ES/neural nets Dear Larry, I strongly recommend you contact A. Robert Spitzer, M.D. c/o University Affiliated Neurologists, P.C.; University Health Center; 4201 St. Antoine-6E; Detroit, MI 48201. He can be reached via telephone at (313)-745-8328. You can reach him via INTERNET using "73226,3642@compuserve.com" although I do not know how often he checks email. I did some work for him on a fascinating neural network which applied quite well to rhythm strips and probably will branch into ECGs as well. He has a related article in SCAMC (1991 I believe). Chuck Kowalewski, D.O. P.S. I have moved to Mississippi to take a job as an ICU director. My new address is 808 W Washington, Greenwood, MS, 38930. Financial pressures have caused me to pursue a clinical career over research. --------------------------------------------------------------------- 15-Feb-93 9:11:08-GMT,2549;000000000000 From: Rob Harrison Date: 15 Feb 93 09:09:13 GMT Subject: Re: Cardiology applications of ES/neural nets Dear Larry Myself and two colleagues have been working on the use of neural networks for the early diagnosis of myocardial infarction. The aim of the work has been to provide decision support during the time between the onset of symptoms and confirmatory tests. For instance, in Edinburgh, where we have been running some trials, the mean time to admission is 3 hours post-onset. Even rapid CKMB mass assay is not too sensitive in this period and is not in general use anyway. Our prototype system, based on a multi-layer Perceptron (Back Propagation Net) has achieved sensitivity, specificity and accuracy of 88% in the lab (ref 1). Futhermore, the network may have a more significant role for atypical patients (ref 2). The system is now undergoing trials in Accident and Emergency Departments in four UK hospitals. We expect results in 18 months. Please contact me if you want to discuss further. Also, I would be very pleased to receive a copy of your talk and/or a digest of the information you pick up on this trawl. It would be very interesting to know who is doing what. Refs 1. R F Harrison, S J Marshall and R L Kennedy (1991) Proc. Eur. Soc. AI in Med. 3rd European Conf. on AI in Med. (AIME), Maastricht, 119-128 A connectionist aid to the early diagnosis of myocardial infarction. 2. R L Kennedy R F Harrison S J Marshall and C A Hardisty (1991) Q J Med. v80 788-789 Analysis of clinical and electrocardiographic data from patients with acute chest pain using a neurocomputer.Robert F Harrison Department of Automatic Control and Systems Engineering University of Sheffield PO Box 600 Mappin Street Sheffield S1 4DU England Tel: +44 (0)742 825139 Fax: +44 (0)742 780409 EMail: R.F.Harrison@uk.ac.sheffield (on JANET) --------------------------------------------------------------------- 15-Feb-93 13:38:12-GMT,2243;000000000000 Date: Mon, 15 Feb 93 13:37:46-0100 From: delaere@esat.kuleuven.ac.be Subject: Cardiology applications of Expert Systems (KULeuven) In respons to the mail you sent to prof. Suetens, I can give you the following information about some projects we have been working on: 1) An expert system for labeling coronary blood vessels, stenosis quantification, 3D-reconstruction from 2 angiographic projections. - "A knowledge-based system for the automatic interpretation of blood vessels on angiogram" C. Smets. Katholieke Universiteit Leuven, Faculteit de geneeskunde/ Toegepaste Wetenschappen PhD thesis, Leuven University Press, 1990. - "A knowledge-based system for the automatic quantification of stenotic lesions on angiograms" D. Delaere, C. Smets, P. Suetens, A. Aubert, F. Van de Werf. Proc. Conf. Computers in Cardiology, pp. 317-320, Chicago, U.S.A., Sep. 23-26, 1990. - "Knowledge-based system for the three-dimensional reconstruction of blood vessels from two angiographic projections" D. Delaere, C. Smets, P. Suetens, G. Marchal, F. Van de Werf. Journal Medical & Biological Engineering & Computing, 29(6), Nov. 1991 pp. NS27-NS36. 2) Contour detection of left ventricle on 2D echocardiograms. - "Automated contour detection of the left ventricle in short axis view and long axis view on 2D echocardiograms" L. Maes, D. Delaere, P. Suetens, A. Aubert, F. Van de Werf. Proc. Conf. Computers in Cardiology, pp. 603-606, Chicago, U.S.A., Sep. 23-26, 1990. Dominique Delaere Interdisciplinary Research Unit of Radiological Imaging (ESAT + Radiology) K.U. Leuven Herestraat 49 3000 Leuven Belgium Tel. +32-16-34 37 49 FAX. +32-16-34 37 69 E-mail: delaere@esat.kuleuven.ac.be --------------------------------------------------------------------- 15-Feb-93 14:07:35-GMT,2836;000000000000 From: ne2@prism.gatech.edu (EZQUERRA, NORBERTO F.) Date: Mon, 15 Feb 93 7:55:54 EST Hello, Larry. I believe you remember me from previous SCAMC meetings and discussions. I saw your "call for information" recently, and wanted to respond. Some of the info that follows may be known to you, in which case pl forgive me. For the last several years, we have been developing, implementing, and testing a methodology for interpreting myocardial perfusion imagery. The methodology consists of a novel approach combining knowledge-based, connectionists, computer vision, and probabilistic methods, to process and interpret SPECT imagery in conjunction with other (non-image) patient-specific information (e.g., EKG info, symptom info, and other personal data). The prototype system is called PERFEX (for perfusion expert) and has evolved in a number of architectures and languages; presently it is implemented in an object-oriented environment on UNIX machines running on X-windows. Among the characteristics of PERFEX: it includes temporal reasoning for addressing the information obtained at different times (i.e., the scans obtained when the patient is stressed, then later when s/he is at rest), probabilistic reasoning to estimate the pre-test likelihood of coronary artery disease, and connectionist techniques to assist in recognizing patterns of myocardial thickening. In addition, it contains a relatively large and robust knowledge base that attempts to represent the visual reasoning associated with interpreting the imagery. PERFEX will be described in an upcoming issue of Expert Systems With Applications, devoted to medical ESs. The sytem is designed with a graphical user interface that supports interactivity throughout the decision-making process. The project has been funded by the NLM of NIH, and is done in collaboration between Georgia Tech and Emory University; I'm the PI, and Dr. Ernie Garcia is the Emory PI. If you would like additional info, pl let me know. Good to hear from you. Best regards, Norberto ------------------------------------------------------------------------------- | NORBERTO EZQUERRA, PhD. | Email : norberto@cc.gatech.edu | | Associate Professor | Phone : (404)-853-9173 | | College of Computing, | Sec. : (404)-853-0672 | | Georgia Tech., Atlanta, GA - 30332 | FAX : (404)-853-0673 | ------------------------------------------------------------------------------- --------------------------------------------------------------------- 16-Feb-93 17:16:11-GMT,805;000000000000 Date: Tue, 16 Feb 1993 10:09:08 -0700 From: crr@cogsci.psych.utah.edu Subject: cardio applications I have an application that is in press (Neural Computation, Vol 5 No.3 I believe) which learns to interpret myocardial scintigrams. Here's how to get the paper from the neuroprose site. unix> ftp cheops.cis.ohio-state.edu Connected to cheops.cis.ohio-state.edu. 220 cheops.cis.ohio-state.edu FTP server ready. Name: anonymous 331 Guest login ok, send ident as password. Password: {your address} 230 Guest login ok, access restrictions apply. ftp> binary 200 Type set to I. ftp> cd pub/neuroprose 250 CWD command successful. ftp> get rosenberg.scintigrams.ps.Z 200 PORT command successful. 150 Opening BINARY mode data connection for rosenberg.scintigrams.ps.Z 226 Transfer complete. 100000 bytes sent in 3.14159 seconds ftp> quit 221 Goodbye unix> uncompress rosenberg.scintigrams.ps.Z unix> lpr rosenberg.scintigrams.ps (or however you print postscript It is the not-quite-up-to-date version of the paper that is coming out in Neurocomputation next month. But the changes are minor. Charlie --------------------------------------------------------------------- 18-Feb-93 15:37:25-GMT,1696;000000000000 Date: Thu, 18 Feb 1993 16:34:52 +0100 From: cagnoni@cobra.ing.unifi.it Subject: Cardiology applications of ES/neural networks Dear Dr. Widman, let me apologize for being a little self-celebrating. I am a Ph. D. student at the University of Florence, working with Prof. G. Valli on AI applications to medical images and signals. One of our fields of interest is the long-term analysis of blood-pressure data. On this subject we have published, in 1991, the paper "Neural network expert system for diagnosing and treating hypertension", which was published in March 1991 issue of IEEE Computer Magazine. We have then presented the evolution of our work in Venice at 'Computers in Cardiology' in September 1991. The Proceedings of this conference are available from IEEE CS Press. I hope our work can be of some help to you in preparing your speech. I will be glad to give you any further detail you might need and, if you are interested, to send you reprints of the papers. Wishing you the greatest success with your lecture, Stefano Cagnoni Stefano Cagnoni Dept. of Electronic Engineering University of Florence I-50139 Florence (Italy) Tel. +39-55-4796378 FAX +39-55-494569 Email cagnoni@cobra.ing.unifi.it --------------------------------------------------------------------- 23-Feb-93 22:43:31-GMT,1429;000000000000 Date: Tue, 23 Feb 93 14:41:50 PST From: sirothe@srv.PacBell.COM (Sheldon Rothenberg) Subject: Re: Cardiology applications of ES/neural nets Am curious as to any thoughts on your part on why there are such few cardiology applications. My limited experience, 2 years as a health counselor ina heart attack prevention program, suggests that many who feel that nutritionists, cardiologists, or psychologists make them uncomfortable might value the anonymity of interacting with informational, advisory systems. From the cardiologists' perspective, I am wondering how much credence a system might hold on topics such as. let's say drug interactions of anti- hypertensives and other medications, therapeutic protocols, or other topics. Of course such a system would need to be updated pretty regularly for it to be trusted.... Shelley --------------------------------------------------------------------- 24-Feb-93 15:04:43-GMT,6135;000000000000 Date: Wed, 24 Feb 93 10:05:02 EST From: jsearle@bmemail.MCG.PeachNet.EDU (John R Searle) Subject: Cardiology ANN Apps Hello Larry, In response to your query about neural network applications in cardiology, here is some info about one application here at the Medical College of Georgia by Weiqun Yang and a few references he gave me. Weiqun's application is the interpretation of high resolution ECG's in the time domain. Specifically he is looking at improving interpretation of low level components of the terminal QRS complex. This analysis of a single beat is difficult to make because the experts have only 3 or 4 parameters to look at. In the neural net application, the last 50 ms of signal are sampled at 1 ms intervals (50 samples) and are input to the ANN. The network was trained with 40 normal beats and 30 beats of clinically verified ventricular tachycardia. Thus the ANN can classify beats as either normal or ventricular tachycardia. The network was tested with 10 additional normal and 10 additional ventricular tachycardia beats and the results seem comparable to an expert's evaluation. Not enough cases have been run through the network to make a statistically valid evaluation. This work has not been published yet. Weiqun works in collaboration with Nancy Flowers, M.D. Here are some citations given to me by Weiqun: Xue, Qiuzhen, Yu Hen Hu, Willis J Tompkins, "Neural-network-based adaptive matched filtering for QRS detection," IEEE Transactions on Biomedical Engineering, 39(4):317-329, 1992. Abstract - We have developed an adaptive matched filtering algorithm based upon an artificial neural network (ANN) for QRS detection. We use an ANN adaptive whitening filter to model the lower frequencies of the ECG which are inherently nonlinear and nonstationary. The residual signal which contains mostly higher frequency QRS complex energy is then passed through a linear matched filter to detect the location of the QRS complex. We developed an algorithm to adaptively update the matched filter template from the detected QRS complex in the ECG signal itself so that the template can be customized to an individual subject. This ANN whitening filter is very effective at removing the time-varying, nonlinear noise characteristic of ECG signals. Using this novel approach, the detection rate for a very noisy patient record in the MIT/BIH arrythmia database is 99.5%, which compares favorably to the 97.5% obtained using a linear adaptive whitening filter and the 96.5% achieved with a bandpass filtering method. Furling, James W, Milton E Dupuy, James A Heinsimer, "Neural network analysis of serial cardiac enzyme data," Am J Clin Pathol 96(1):134-141, 1991. Abstract - There has been a recent resurgence of interest in the study and application of computerized neural networks within the broad field of artificial intelligence. These "intelligent machines" are modeled after biological nervous systems and are fundamentally different from the many computerized expert systems that previously have been introduced as clinical decision-making dads. The authors describe a neural network designed and trained to predict the probability of acute myocardial infarction (AMI) based on the analysis of paired sets of cardiac enzymes. The neural network predicted 24 of 24 (100%) AMIs and 27 of 29 (93%) No-AMIs when compared with a pathologist's interpretation of the patient's laboratory data (P<0.000001). The authors attempted to validate the network's diagnoses by two independent methods. When compared with echocardiogram and EKG for diagnosis of AMI, the neural network agreed with the cardiologist's interpretation in 12 of 14 (86%) AMIs and 1 of 3 (33%) No-AMIs, but the correlation was not statistically significant. Using autopsy outcome for validation, the neural network agreed with the anatomic evidence in 24 of 26 (92%) AMIs and 4 of 6 (67%) No-AMIs (P=0.001). The authors conclude that neural networks can be successfully applied to the analysis of cardiac enzyme data and suggest that broader applications exist within the domain of clinical decision support. Dassen, W R M, R Mulleneers, J Smeets, K Dulk, F Cruz, P Brugada, H J J Wellens, "Self-learning neural networks in electrocardiography," J Electrocardiol 23(suppl): 200-202, ?? No Abstract. Conclusion of paper - The neural network approach can already be used for partial interpretation of the electrocardiogram. It represents a powerful tool for defining and evaluating new criteria. But while building neural systems one should take care not to fall prey to one large pitfall: Neural networks are like small children: For every question they have an answer! Baxt, William G, "Use of an artificial neural network for the diagnosis of myocardial infraction," Annals Internal Medicine 115:843-848, 1991. Conclusion of Abstract - An artificial neural network trained to identify myocardial infarction in adult patients presenting to an emergency department may be a valuable aid to the clinical diagnosis of myocardial infarction; however, this possibility must be confirmed through prospective testing an a larger patient sample. Here is a question for you. Do you know anyone who has used "Outcome Advisor"? As I understand it, this is a commerical neural network capable of self organization or modification. Regards, John Searle __________________________________________________________________________ | John R. Searle, Ph.D. | Email: jsearle@bmemail.MCG.PeachNet.edu | | Biomedical Engineering CN-135 | Office: (706)721-4110 | | Medical College of Georgia | Sec'y: (706)721-3161 | | Augusta, GA 30912-4610 | Fax: (706)721-6277 | |__________________________________________________________________________| --------------------------------------------------------------------- 3-Mar-93 17:54:30-GMT,2957;000000000001 Date: Wed, 3 Mar 1993 18:53:15 +0100 From: Georg Dorffner Subject: Re: Cardiology applications of ES/neural nets Dear Dr. Widman, For the last two, three years we have been working on a cardiological application with neural networks, namely the interpretation of thallium scintigrams with respect to coronary artery disease, with quite good success. 'we' means the Austrian Research Institute for Artificial Intelligence and the Dept. of Cardiology at the Clinic of Internal Medicine of the University of Vienna. The following abstracts and papers have been published: /1/ Dorffner G., Kundrat S., Petta P., Porenta G., Duit J., Sochor H.: Interpretation von segmentalen Thallium-201 Szintigrammen mit neuronalen Netzwerken (abstract), Wiener Klinische Wochenschrift, 15-91, 1991. /2/ Dorffner G., Prem E., Mackinger M., Kundrat S., Petta P., Porenta G., Sochor H.: Experiences with Neural Networks as a Diagnostic Tool in Medical Image Processing, to appear in: 37. Jahrestagung der Dt. ges. fuer Med. Informatik, Biometrie und Epidemologie (GMDS), Proceedings, 1993. /3/ Porenta G., Dorffner G., Schedlmayer J., Sochor H.: Parallel Distributed Processing as a Decision Support Approach in the Analysis of Thallium-201 Scintigrams, in Proc. Computers in Cardiology 1988, Washington, D.C., IEEE, 1988. /4/ Porenta G., Dorffner G., Sochor H.: Computer Assisted Interpretation of Thallium-201 Scintigrams Using the Parallel Distributed Processing Approach, in Proc. 10th Congress of the European Society of Cardiology, Vienna, (Abstract), 1988. /5/ Prem E., Mackinger M., Dorffner G., Porenta G., Sochor H.: Concept Support as a Method for Programming Neural Networks with Symbolic Knowledge, in Ohlbach (ed.): Proceedings of the 16th German AI- Conference (GWAI-92), pp.166-175, Springer Verlag, Berlin- Heidelberg, Lecture Notes on AI, Vol. 671, 1993. We are also in the process of submitting a detailed article to Circulation. If you are interested in more infromation about our research, please let me know. sincerely, Dr. Georg Dorffner Head of the Neural Network Group Austrian Research Institute for Artificial Intelligence Schottengasse 3 A-1010 Vienna, Austria Tel: +43-1-53532810 Fax: +43-1-5320652 email: georg@ai.univie.ac.at .