A Patient Care Digital Library Personalized Search and Summarization over Multimedia Information Kathleen McKeown, Shih-Fu Chang, James Cimino, Paul Clayton, Judith Klavans Columbia University 1 Vision Personalized access to distributed resources information vetting information fusion information understanding Provision of patient-specific information at the point of patient care in terms that can be understood Benefits: awareness of new procedures, increased compliance 2 Features Tailor search, presentation, and summarization of online medical literature Exploit online patient record as user model Patients and health care providers Multimodal access and presentation over multimedia resources 3 Scope of Research “exciting,” “innovative,” “broad,” “ambitious,” “complex” Computer Science and Medical Informatics Links into libraries: CRIA Suite of existing component technologies Five year vision Active partners 4 Continuum Current: existing individual components for multiple applications summarization of terrorist news, image search, categorization of images, multimedia briefings, PATCIS, patient mammography record extraction, video summarization of news, query translation Future: DLI2 enables unification and transformation of components new level of capability integrated in a single application 5 Outline Desmond Jordan Why this research is needed Kathy McKeown An example—integration of components Scope, organization, management Role of partners and new developments Jim Cimino Clinical information infrastructure Kathy McKeown Budget 6 Rounds Patient-centric Current: Access to clinical data Missing: Access to literature that fits patient profile 7 Data follows the patient—why not literature? 8 Ongoing Collaboration: Generation of Multimedia Briefings Voice: . . . . Ms. Jones is an 80 year old, diabetic, hypertensive, female patient of Dr. Smith undergoing CABG. . . . 9 User Study Group Speech, text, and graphics output were useful 10 Steve Feiner: Multimedia Video If I recall correctly, I found the image at the bottom in a we page for patients with CHF. May have something to do with fluid buildup. Physicians: residents learn by comparison Patients: Video helps them understand their diagnosis They like to investigate prognosis 11 Query Steve Feiner: ThisWhy slide differs from the previous one should I prescribe amiodarone? in that it has a real graphical history, although not in the medical domain. It may be better, since it means that all of the material on the slide will be real. WebSeek BTW, this is a “comment”. You can either delete it or make it invisible with the “View” menu “comments” command History Lexical Navigation Key Issues: User Interface Query augmentation using the patient record Congestive Heart Failure (CHF), male, Base Heart Rate (BHR) > 90, history of ischemia Lucent: speech, IBM: query refinement, interface for image/video search 14 Distributed Search Antiarrhythmic Drugs NSF is planning to implement a new electronic project reporting system through which proposers will be able to produce the Results from Prior Support section of the proposal. This new system will be part of the NSF FastLane system and will permit updating of reports on each of the PI's NSF projects and will avoid re-en the normal mechanism for project reporting to NS Documents Amiodarone Amiodarone profile profile profile Amiodarone Treatme CHF NSF is planning to implement a new electronic project reporting system through which proposers will be able to produce the Results from Prior Support section of the proposal. This new system will be part of the NSF FastLane system and will permit updating of reports on each of the PI's NSF projects and will avoid re-entry of information previously provided by the PI, either with the original proposal submission or in updates provided usingting process and become the normal mechanism for project reporting to NS Augmented query b Search Amiodarone CHF male BHR > 90 Echocardiogram Echo 10/15/98 Echo 9/15/98 Patient heart sounds Database of Videos Echo 8/15/98 Echo 7/15/98 Amiodarone CHF male BHR > 90 Journal Articles Methods: All patients had advanced CHF with ... Documents Amiodarone Amiodarone Amiodarone Amiodarone NSF is planning to implement a new electronic project reporting system through which proposers will be able to produce the Results from Prior Support section of the proposal. This new system will be part of the NSF FastLane system and will permit updating of reports on each of the PI's NSF projects and will avoid re-entry of information previously provided by the PI, either with the original proposal submission or in updates provided usingting process and become the normal mechanism for project reporting to NS Categorization & Filtering NSF is planning to implement a new electronic project reporting system through which proposers will be able to produce the Results from Prior Support section of the proposal. This new system will be part of the NSF FastLane system and will permit updating of reports on each of the PI's NSF projects and will avoid re-entry of information previously provided by the PI, either with the original proposal submission or in updates provided usingting process and become the normal mechanism for project reporting to NS Prognosis Article NSF is planning to implement a new electronic project reporting system through which proposers will be able to produce the Results from Prior Support section of the proposal. This new system will be part of the NSF FastLane system and will permit updating of reports on each of the PI's NSF projects and will avoid re-entry of information previously provided by the PI, either with the original proposal submission or in updates provided usingting process and become the normal mechanism for project reporting to NS Prognosis Article NSF is planning to implement a new electronic project reporting system through which proposers will be able to produce the Results from Prior Support section of the proposal. This new system will be part of the NSF FastLane system and will permit updating of reports on each of the PI's NSF projects and will avoid re-entry of information previously provided by the PI, either with the original proposal submission or in updates provided usingting process and become the normal mechanism for project reporting to NS Diagnosis Article . . . Target Summaries Prognosis articles: amiodarone patient, BHR > 90 A significant reduction in sudden death was observed in patients with a BHR > 90 beats/min treated with amiodarone pacemaker patient, BHR 90 In patients with a BHR 90 beats/min, amiodarone-treated patients had a mortality rate similar to those not treated with amiodarone. One article found that pacing fails to improve hemodynamic function, while another showed a significantly better quality of life. PERSIVAL Management Structure Technology Search Gravano User Interface Johnson Summarization McKeown Project Leader McKeown Clinical Infrastructure Cimino Evaluation Industry Partners Klavans Patel Content and Library Team Klavans PERSIVAL Management Structure Technology Search Gravano User Interface Johnson Summarization McKeown Project Leader McKeown K. McKeown Clinical Infrastructure Cimino Evaluation Industry Partners Klavans Patel Content and Library Team Klavans PERSIVAL Management Structure Technology Search Gravano User Interface Johnson Summarization McKeown Project Leader McKeown J. Klavans Clinical Infrastructure Cimino Evaluation Industry Partners Klavans Patel Content and Library Team Klavans PERSIVAL Management Structure Technology Search L. Gravano Gravano User Interface Johnson Summarization McKeown S. Johnson Project Leader McKeown Clinical Infrastructure J. Cimino Cimino Evaluation Industry Partners Klavans Patel Content and Library Team Klavans S.-F. Chang PERSIVAL Management Structure Technology Search S. Johnson Gravano User Interface Johnson Summarization S. Feiner McKeown Project Leader McKeown Clinical Infrastructure Cimino Evaluation Industry Partners Klavans Patel Content and Library Team Klavans S.-F. Chang PERSIVAL Management Structure Technology Search Gravano User Interface Johnson K. McKeown Summarization McKeown Project Leader McKeown Clinical Infrastructure Cimino J. Klavans Evaluation Industry Partners Klavans Patel Content and Library Team Klavans V. Hatzivassiloglou PERSIVAL Management Structure Technology Search J. Cimino Gravano User Interface Johnson Summarization McKeown D. Jordan Project Leader McKeown Clinical Infrastructure Cimino C. Friedman Evaluation Industry Partners Klavans Patel Content and Library Team Klavans S. Johnson PERSIVAL Management Structure Technology Search Gravano User Interface Johnson V. Patel Summarization McKeown Project Leader McKeown Clinical Infrastructure Cimino J. Cimino Evaluation Industry Partners Klavans Patel Content and Library Team Klavans D. Jordan PERSIVAL Management Structure Technology Search J. Klavans Gravano User Interface Johnson Summarization S. Jacobson McKeown Project Leader McKeown Clinical Infrastructure P. Molholt Cimino Evaluation Industry Partners Klavans Patel Content and Library Team Klavans E. La Rue Role of Technology Partners Lucent: Spoken Language Dialogue Systems (speaker independent) IBM: Query refinement, evaluation, text summarization, interface for image/video search GE: Concept based retrieval 32 New Partners and Changes Siemens Corporate Research: grant for video summarization integrating text/image, plus researcher time $68,680/year AT&T: grant to build our networking infrastructure $50,000 Welch Allyn: information and equipment, Medical Director’s time, cash $35,000 GE Medical Systems: commitment to commercialization of results 33 Steve Feiner: IBM is more than just a software company; ditto Lucent, GE, etc. Why not just group companies together? Partners Partner General Electric Team IBM Companies Lucent Siemens Libraries & New York Public Museums Library MOMA Donating Bell Atlantic Companies Intel AT&T Contribution Software Researcher Software Researcher Software Researcher Student Support Grant Researcher Student Support Consumer Health Information Specialists Grant, Image Processing Technology Authentication Services Computer Equipment Grant, Network Infrastructure Value $200,000 $15,000/year $50,000 $110,000/year $25,000 $125,000/year $201,286 over 5 years $25,000/year $34,000/year $9,680/year $10,000 $115,000 $209,154 over 5 years $50,000 34 Partners Partner Healthwise Content Lion Reef Providers OVID Technology Univ. of CA San Francisco Welch Allyn Contribution Value Consumer Health & $10,000/year Software Images & Animation of $30,000 Heart Scientific Biomedical Data $15,000/year & Software Medical Images Information and equipment Medical Director’s time Unrestricted funds $20,000 $10,000 $5,000 35 Clinical Information Systems Clinical repository of coded and text data Medical Entities Dictionary server Web-based access for clinicians (WebCIS) Web-based access for patient (PatCIS) Infrastructure handles security, application tracking, and evaluation Architecture handles integration of components 36 Input for PERSIVAL Coded data Natural language processing Medical Entities Dictionary UMLS 37 Web/PatCIS and Digital Library Infobutton Information Resource WebCIS/ PatCIS CIS MED 42 Web/PatCIS and Digital Library Infobutton Information Resource Query Interface PERSIVAL WebCIS/ PatCIS CIS MED 43 Users Clinical users Columbia-Presbyterian Medical Center New York Hospital 37 affiliates Patients Heart failure Home diabetes telemonitoring High risk for heart attack Cancer patients 44 Evaluation and User-Centered Design Usability engineering Evaluation of system components: cognitive and quantitative in situ and lab video recordings Subjective assessment of search plus recall, precision Survivability end-to-end evaluation, reduction of manual intervention, training, university support of applications 45 Summary Tailor search, presentation, and summarization of online medical literature Exploit online patient record as user model Patients and health care providers Multimodal access and presentation over multimedia resources 46