18
Diagnostic Cockpit of the Future workshop 16-17 May 2018 Gaithersburg, MD Summary A two-day workshop, sponsored by the Academy of Radiology and Biomedical Imaging Research was held on the campus of the National Institute of Standards and Technology (NIST) in Gaithersburg, Maryland. Workshop participants included representatives from industry, academia, government and members of the Coalition for Imaging and Bioengineering Research (CIBR). The goal of the workshop was to enable the creation of a “diagnostic cockpit” where medical experts have access to the data needed to diagnose and treat patients while artificial intelligence (AI) analyzes that data and provides decision support. Three broad themes that were discussed included: 1) case studies (hepatocellular cancer (HCC), cardiac disease, prostate cancer), 2) core functional requirements for the diagnostic cockpit and 3) initiatives. A description of the Academy-NIST Workshop and the workshop goals for the Diagnostic Cockpit of the Future can be found online: http://bit.ly/AcademyNIST . Below is a summary of breakout sessions and workshop wrap-up. Topics Summary.............................................................. 1 Environmental factors, billing/patient/workflow/diagnostician perspective.......................................................... 2 Case Study: Cardiovascular Disease..................................2 Current state of data streams........................................4 Five things of Dx Cockpit:..........................................4 Steps toward Diagnostic Cockpit.....................................5

Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

  • Upload
    ngongoc

  • View
    214

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

Diagnostic Cockpit of the Future workshop

16-17 May 2018

Gaithersburg, MD

SummaryA two-day workshop, sponsored by the Academy of Radiology and Biomedical Imaging Research was held on the campus of the National Institute of Standards and Technology (NIST) in Gaithersburg, Maryland. Workshop participants included representatives from industry, academia, government and members of the Coalition for Imaging and Bioengineering Research (CIBR).

The goal of the workshop was to enable the creation of a “diagnostic cockpit” where medical experts have access to the data needed to diagnose and treat patients while artificial intelligence (AI) analyzes that data and provides decision support.

Three broad themes that were discussed included:

1) case studies (hepatocellular cancer (HCC), cardiac disease, prostate cancer),

2) core functional requirements for the diagnostic cockpit and

3) initiatives.

A description of the Academy-NIST Workshop and the workshop goals for the Diagnostic Cockpit of the Future can be found online: http://bit.ly/AcademyNIST. Below is a summary of breakout sessions and workshop wrap-up.

TopicsSummary.....................................................................................................................................................1

Environmental factors, billing/patient/workflow/diagnostician perspective..............................................2

Case Study: Cardiovascular Disease.........................................................................................................2

Current state of data streams......................................................................................................................4

Five things of Dx Cockpit:........................................................................................................................4

Steps toward Diagnostic Cockpit.............................................................................................................5

Use case: Myocardial ischemia................................................................................................................5

Standards.................................................................................................................................................6

Standardization...........................................................................................................................................7

Data needed..........................................................................................................................................11

Input and components of Diagnostic Cockpit........................................................................................16

Day 2 Summaries.......................................................................................................................................17

Page 2: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

Environmental factors, billing/patient/workflow/diagnostician perspective

Diagnostic Cockpit Breakout Session Room D

16 May 2018

Standardization of Radiology workflow. Diagnostician and patient point of view (POV) needed. Is this something like DICOM or MQSA? ACR practice guidelines – [Can these be used]?

Case Study: Cardiovascular Disease EKG can be read by AI Atlas of nuclear cardiology is accepted standard (map)

Figure: PET map provides quantitative map of blood flow.

Aggregating data may be first step toward Diagnostic Cockpito E.g., integration of EPIC and PACS.

Cockpit may be physician-specialty specific, patient-specific, Workflow? These includes things like display and data elements. Incentives? Who will pay for it? U. MD. Paid $500M for EPIC

o Decision support support is Zero

Aggregating data may be first step toward Diagnostic Cockpit(e.g., integration of EPIC and PACS.

Page 3: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

Allocate 1% or 2% for decision support Industry partner needed. CTO and CFO does not understand importance. NEMA came in with DICOM as a neutral party. We need something analogous.

Page 4: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

Current state of data streams and how do we connect them?

Diagnostic Breakout session Room ARich Mather, moderator

1. Earlier group session discussed: X rads, birads, li rads, Lung Rads, etc. 2. Lab measures3. Claim / ICD 10 codes

Potential test cases and how to leverage them (data streams) Prostate, breast, cystic fibrosis What are barriers to aggregating data? How do we define standards (to detect cancer, for example) Is there a killer use-case? Data aggregator and dashboard [first step]

Five things of Dx Cockpit:1. Assist image interpretation2. Assist physician3. Patient management options4. ? 5. ?

Minimal [information] needed for data aggregatoro Ontologies, data elements, etc. need for data aggregatoro How do you measure impact? Influence a decision.o Oncotype Dx – a molecular assay

Possible use case – Lung nodule follow-up Ontologies and reporting – Tragic missing link, even when used all reports Patient-specific management options should be considered– e.g., breast cancer of marathon

runner treated differently from other patients (Mitch Schnall)

Page 5: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

Steps toward Diagnostic Cockpit

1. Data Aggregator that enables physician to make the correct decision faster. 2. Plug into data from commercial and academic sources3. DEFINE USE CASE4. Develop analytic layer5. Learning health system

Confidence of interpretation is critical Result is more critical – How do you measure the quality of radiologist? Downstream testing needed Collect data from Radiologists to make report RADS to report to RADS RADS is a datastream, text reports, etc.

Use case: Myocardial ischemia

[SYMPTOMS or clinical scenario] Coronary artery disease – myocardial ischemia

Clinical scenario – not a specific test (PET image of perfusion) is needed Drives an intervention – ANGIOGRAM Becomes a tool for the Radiologist Engage ACC (society) CAD RADS is targeted at coronary artery disease.

Define user experience FIRST. Define task, test it on users Trying to get everyone to build “one” [Diagnostic Cockpit] Forcing function – public private partnership Neutral group could create the prototype (e.g. Lincoln lab)

Data Aggregator enables physician to make the correct decision faster.

The myocardial ischemia is a clinical scenario (use case) that drives an intervention (angiography).

Page 6: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

We need to create something [report?] that can be used across the board (by industry competitors, for example)

Develop data standards, scenarios, each developer must pass muster CT perfusion vs. MR perfusion are widely varying

Standards Standard APIs for an App Framework

o This [kind of thing] could enable Analytic Apps Digital reference objects are a great [standard] for this kind of development Contextualize contextualization – It takes 30 secs to look at picture, 30 minutes to look at record

o Learn this from a Radiologist. E.g., Pulmologist says, “I don’t consider this person a radiologist, I consider him

a chest physician.”

Robot physician or robot radiologist? o Robot provides context

ID findings, integrate assessments AI image interpretation will replace boring, dull [repetitive tasks]

o E.g., Detecting pulmonary nodules. Radiologists HATE reading pulmonary nodules. [This is a dull, repetitive, task.

20 years from now what do diagnostics look like? Integration of information. Some entity owns the process. This does not exist now. Diagnostic specialty integrates clinical med into infrastructure. Farm out contextualization to a robotic aggregator. IT Analytics backbone, Who manages in that health system? “Diagnostic Department.” Global Advisory Board needed for ??? A Request for Information (RFI) is needed. What is our groups “take home message?”

NIST [example] First Net – sold bandwidth for public safety communications. WHAT ARE THE NEEDS THAT ARE NOT ALREADY SUPPORTED with PI Pilot or feasibility study is needed. Public private partnership around one domain areas (for 3 killer apps).

Need AREA [of focus] WHAT IS THE MISSING AREA? [Government] Request for Applications (RFA) is not needed. The National Science and Technology Council (NSTC) within OSTP DOD, NSF, NIH, etc. can

help Example: Precision Medicine, Cancer Moonshot are NIH Common Fund initiatives

Page 7: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

StandardizationBreakout Session Room B

Page 8: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference
Page 9: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference
Page 10: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference
Page 11: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

Data needed

Page 12: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference
Page 13: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference
Page 14: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference
Page 15: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference
Page 16: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

Input and components of Diagnostic Cockpit

Page 17: Summary · Web viewCT perfusion vs. MR perfusion are widely varying Standards Standard APIs for an App Framework This [kind of thing] could enable Analytic Apps Digital reference

Day 2 SummariesRich Mather (Session A)

Liver and cardiac were designated targets Recommend acquiring 100 datasets [patients] from 10x sites

Bram Stolk (GE) Session B

Create compendium of standards from existing standards – ACR Registry of Standards and Interoperability

Use “connectathons: at technical meeting (HIMSS and SIIM)

Broad themes

1. Case Studiesa. Hepatocellular cancer (HCC), cardiac disease, prostate cancer

2. Core functional requirements [for DxCP]a. LOINC (e.g.) addresses priors and interoperabilityb. UMLS and FIRE tags and attributes can be used in industry-specific flavored structured

reportc. API (Application Programming Interface)d. Cockpit could be anatomy [atlas] basede. Functional specs are done [deal]. Companies are doing this – per Bram Stolk

3. Initiatives [Funding]a. Scope $10M - $100Mb. NIH P41 - $20M over 15 yearsc. OSTP Agency Directors (e.g., NIH, NSF, DOD) National Initiative (like Cancer

moonshot) (e.g., $100M to NIH Common Fund (Director’s request)