Title Att3 Presentation TG Histo Purpose Discussion Contact Frederick Klauschen Department of Pathology LMU Munich amp Charité Berlin Email FrederickKlauschenmedunimuenchende ID: 927653
Download Presentation The PPT/PDF document "Source: TG- Histo Topic Driver" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Source:
TG-Histo Topic DriverTitle:Att.3 – Presentation (TG-Histo)Purpose:DiscussionContact:Frederick KlauschenDepartment of PathologyLMU Munich & Charité BerlinE-mail: Frederick.Klauschen@med.uni-muenchen.de Abstract:This PPT summarizes the status of work within TG-Histo, for presentation and discussion during the meeting.
FGAI4H-L-013-A03
E-meeting, 19-21 May 2021
Slide2WHO/ITU FG AI4HEALTH
20. 5. 2021Frederick KlauschenDepartment of PathologyLMU Munich & Charité BerlinTopic GroupHistopathology
Slide3Histological slide
Microscopicdiagnostics
Slide4Manual
evaluation
Slide5Slide6Identify
& classifycancer!EstimateImmune cells!
Slide7Artificial
Intelligence in Diagnostics:PATHOLOGISTS IN DANGER?PATIENTSNeed for validation and benchmarking of
AI in medicine
!
Slide8ITU/WHO Focus Group AI for Health
Topic Group HistopathologyTopic group dedicated to benchmarking AI approaches in histopathologyFirst use case: Detection of breast cancer cells and tumor-infiltrating lymphocytesDefine what should be annotated and how Define criteria for benchmarkingProvide server infrastructure to perform benchmarking
Slide9Annotation of the histopathology images
Specifications:Digitized histological slides in standard stainingComprehensive tissue component annotations:cancer tissuemultiple subtypes focus on NST (no-special-type) and invasive-lobular breast cancernormal tissuenormal breast gland and duct epitheliumconnective tissue (fibers
, cells)
fatty tissue
,
bone tissue, nerves
blood and lymphatic vessels
immune system
Lymphocytes, plasma cells
Granulocytes
,
monocytes/macrophages
necrotic tissue
artifacts
Background
Positive
and
negative
annotations
Slide103 Benchmarking
true
positive,
true
negative,
precision
Slide11Benchmarking pipeline
ITUServer(undisclosed data)Benchmarking Data Annotation
AI developer
(
approach
trained
with
own
data
)
Submit
data
Submit
algorithm
Report
results
WHO/ITU
publication
public
example
data
Slide12Benchmarking breast cancer cell detection
HHIServerBerlinBenchmarking Data Annotation AI developer(approach
trained
with
own
data
)
Submit
data
Submit
algorithm
Report
results
Cancer
cell
detection
algorithm
,
Prof. Alex Binder,
Singapore
Breast
cancer
data
set
, 50
patients
approx
. 10k
annotations
Binder_A_2019:
tp
=0.91,
tn
=0.88
Prof. Dr. Alexander Binder
Singapore
University
of
Technology
and
Design (SUTD)
Singapore
,
now
Oslo Univ.
Slide13image data 3000x3000 at 400x
consensus annotations by two pathologists5 exemplary images made available with test annotations to provide overview of data 50 annotated images not public, available only for benchmarking on WHO/ITU servers with>10k annotations Provision of test and benchmarking images
Slide14Additional
cases/tumor types (tissue microarray and whole slide):100 cases NSCLC (non-small cell lung cancer)100 cases breast cancer Different whole slide scanners:3DHistech, Philips, LeicaMore pathologists Currently national and international outreach Preferably advertising through FGAI4HEALTH Websiteregister at f.klauschen@lmu.deAccess through annotation portal at http://annotation.network
Extension of data sets:
Slide15Webportal at annotation.network
Slide16Webportal at annotation.network
Slide17Design of a ITU/WHO validation data set
Multicentric data set:Different labs/stainingsDifferent scannersAnnotation.NetworkAnnotations by pathologists
from
various
institutions
/countries
FG AI4HEALTH Server (undisclosed data)
Slide18Development of accreditation standards
for digital pathology and artificial intelligence1st committee meeting took place in MayMembership in German Accredidation Board Subcommittee for Digital Pathology Comments/participation requests:Frederick Klauschen f.klauschen@lmu.de