Physician Assistant Artificial Intelligence
Reference System
AM Mohan Rao
Umashankar Adi Kotturu
A. Sri Kailash
www.ai-med.in/pairs/
Pain factors
• Misdiagnosis
– Common vs rare disease
• Missed diagnosis
– Errors in clinical data
• Delayed diagnosis
- Complex case
• Treatment costs
- Unnecessary testing
• Drugs
- Side effects
• Reasoning
-Uncertainty of clinical data
• Perception
-Vast domain knowledge
• Diagnostic bias
-Experience
• Inference
-Infection? Neoplasia?
• Training
-Latest advances
Patient Doctor
2
Solutions
• Internet
- Google search
- Social media
• Colleagues
- Seminars, discussions
• Journals, books
-Access to quality information
• Timely advise
-Emergencies
• Resources
-Limitations in remote settings
• Web application
-Bi-layered Google search
-WhatsApp Messages
• Mobile app
-Artificial intelligence
• Database
-SNOMED CT 426 000 terms
• Diagnostic Decision
Support (DDS)
-Logic & probability
• Natural Language
Processing (NLP)
-Text & XML records
General Ai-med.in
3
Physician Assistant Artificial Intelligence
Reference System(PAIRS)
• Web application
-Bi-layered Google Search, DDS
• Mobile app
-Android and iPhone, NLP & DDS
• NLP
- Based on SNOMED CT algorithm
• DDS
-Based on Bayesian method
• Database
-PAIRS specific: 18 397 for 485 diseases and 1964 findings
-SNOMED CT: 426 000 terms, 5190055 relationships
4
Search Engine
PAIRS Google
Bi-layered Single layered
SNOMED CT algorithm + Google search Google search alone
Pathophysiological + Computer based Computer algorithm alone
Context based Word based
Limited relevant search Exhaustive irrelevant search
5
Google Search
6
PAIRS Search
7
PAIRS Google Search
8
Diagnostic engine
• Inference engine
-3 levels
• Feature given disease
- (a).concomitant in assertion & negation
- (b).concomitant in assertion only
- (c).concomitant in negation only
• Ontological class
-Both system and organ are shared
-Only system is shared
-Neither system nor organ are shared
• Word vectors
-Medical text corpus 50 million words
• Bayesian probability
-Lower bounds
• Feature given disease
-(a). Biopsy (b). Deep tendon
reflexes brisk (c). Loss of
tendon reflexes
• Ontological class
- of disease feature links
• Word vectors
- 485 diseases, 1964 findings
9
PAIRS DDS
Patient data entry: directly or by file (txt or xml)
10
PAIRS DDS: Diagnostic types
11
PAIRS DDS
Diagnostic output for different types
12
Marketing | strategy
• Licensing
-Hospitals
-Medical colleges
-Residents and Medical students
-Pharmaceutical companies
-Telemedicine
• Advertisement
-Drugs and brands
-Side effects
• Development
-Database
-Diagnosis
• Evaluation
-tertiary hospital
• Publications
-PAIRS evaluations
13
Team
• AM Mohan Rao
-Full time employee
-Worked in Nobel laureates
environment at a top notch
research institute in US
-Committed to work for
breakthrough technology in
Medicine. 35 years
experience.
• Dr. N.S.N. Rao -Professor of Pediatrics
• Dr. P.N. Rao -Gastroenterologist
• Dr. Ravi Kalaputapu –Strategic advisor
• Uma Shankar Adi
-Entrepreneur and evangelist
-20 years experience
in research and development
• Sri Kailash
Design engineer
• Anand Pothapragada
Web site and MySql
Main Innovator Project associates
Advisors
14
Financials |Projections
• Product development
completed
• Require evaluations in
hospital for 2-4 months
• Licensing product to
corporate hospitals
• Brands and drug ads to
pharma companies
• Money needed for
office set up
• Build up a team to
include full time
doctors, software and
marketing
professionals.
15
References
• 1. Subsumptive reflection in SNOMED CT: a large
description logic-based terminology for diagnosis
http://arxiv.org/abs/1512.03516
• 2. Using SNOMED CT concepts for PAIRS
https://www.researchgate.net/publication/221426464_Using_S
NOMED_CT_concepts_for_PAIRS
• 3. And now, artificial intelligence as a medical tool
http://www.thehindu.com/2003/06/09/stories/200306
0903150500.htm
16
Thank you
17

AI-MED

  • 1.
    Physician Assistant ArtificialIntelligence Reference System AM Mohan Rao Umashankar Adi Kotturu A. Sri Kailash www.ai-med.in/pairs/
  • 2.
    Pain factors • Misdiagnosis –Common vs rare disease • Missed diagnosis – Errors in clinical data • Delayed diagnosis - Complex case • Treatment costs - Unnecessary testing • Drugs - Side effects • Reasoning -Uncertainty of clinical data • Perception -Vast domain knowledge • Diagnostic bias -Experience • Inference -Infection? Neoplasia? • Training -Latest advances Patient Doctor 2
  • 3.
    Solutions • Internet - Googlesearch - Social media • Colleagues - Seminars, discussions • Journals, books -Access to quality information • Timely advise -Emergencies • Resources -Limitations in remote settings • Web application -Bi-layered Google search -WhatsApp Messages • Mobile app -Artificial intelligence • Database -SNOMED CT 426 000 terms • Diagnostic Decision Support (DDS) -Logic & probability • Natural Language Processing (NLP) -Text & XML records General Ai-med.in 3
  • 4.
    Physician Assistant ArtificialIntelligence Reference System(PAIRS) • Web application -Bi-layered Google Search, DDS • Mobile app -Android and iPhone, NLP & DDS • NLP - Based on SNOMED CT algorithm • DDS -Based on Bayesian method • Database -PAIRS specific: 18 397 for 485 diseases and 1964 findings -SNOMED CT: 426 000 terms, 5190055 relationships 4
  • 5.
    Search Engine PAIRS Google Bi-layeredSingle layered SNOMED CT algorithm + Google search Google search alone Pathophysiological + Computer based Computer algorithm alone Context based Word based Limited relevant search Exhaustive irrelevant search 5
  • 6.
  • 7.
  • 8.
  • 9.
    Diagnostic engine • Inferenceengine -3 levels • Feature given disease - (a).concomitant in assertion & negation - (b).concomitant in assertion only - (c).concomitant in negation only • Ontological class -Both system and organ are shared -Only system is shared -Neither system nor organ are shared • Word vectors -Medical text corpus 50 million words • Bayesian probability -Lower bounds • Feature given disease -(a). Biopsy (b). Deep tendon reflexes brisk (c). Loss of tendon reflexes • Ontological class - of disease feature links • Word vectors - 485 diseases, 1964 findings 9
  • 10.
    PAIRS DDS Patient dataentry: directly or by file (txt or xml) 10
  • 11.
  • 12.
    PAIRS DDS Diagnostic outputfor different types 12
  • 13.
    Marketing | strategy •Licensing -Hospitals -Medical colleges -Residents and Medical students -Pharmaceutical companies -Telemedicine • Advertisement -Drugs and brands -Side effects • Development -Database -Diagnosis • Evaluation -tertiary hospital • Publications -PAIRS evaluations 13
  • 14.
    Team • AM MohanRao -Full time employee -Worked in Nobel laureates environment at a top notch research institute in US -Committed to work for breakthrough technology in Medicine. 35 years experience. • Dr. N.S.N. Rao -Professor of Pediatrics • Dr. P.N. Rao -Gastroenterologist • Dr. Ravi Kalaputapu –Strategic advisor • Uma Shankar Adi -Entrepreneur and evangelist -20 years experience in research and development • Sri Kailash Design engineer • Anand Pothapragada Web site and MySql Main Innovator Project associates Advisors 14
  • 15.
    Financials |Projections • Productdevelopment completed • Require evaluations in hospital for 2-4 months • Licensing product to corporate hospitals • Brands and drug ads to pharma companies • Money needed for office set up • Build up a team to include full time doctors, software and marketing professionals. 15
  • 16.
    References • 1. Subsumptivereflection in SNOMED CT: a large description logic-based terminology for diagnosis http://arxiv.org/abs/1512.03516 • 2. Using SNOMED CT concepts for PAIRS https://www.researchgate.net/publication/221426464_Using_S NOMED_CT_concepts_for_PAIRS • 3. And now, artificial intelligence as a medical tool http://www.thehindu.com/2003/06/09/stories/200306 0903150500.htm 16
  • 17.