high volume slide scanning architecture and applications dr. andré huisman department of pathology...

22
High Volume Slide Scanning Architecture and Applications

Upload: sybil-maxwell

Post on 24-Dec-2015

221 views

Category:

Documents


1 download

TRANSCRIPT

High Volume Slide Scanning

Architecture and Applications

Department of pathology UMCU

• UMC Utrecht

• >1,000 beds

• >10,000 employees

• Department of pathology:

• 20.000 surgical pathology cases

• 156.000 glass slides (histology, cytology, IHC)

• 15 pathologists, 10 residents

Digital pathology - advantages

• Digital Archiving

• Instant access from multiple locations by multiple people

• No searching for slides

• Constant quality

• Telepathology

• Consultations, revisions and panels

• Education

• Research

• Automated image processing

Project background

Clinico pathological conferences:

• 900 meetings every year

• No (multi headed) microscope needed

• Quicker preparation of meetings

• No retrieval of glass slides from archive

Aim (2007)

Digitize all diagnostic slides we have

(prospectively)

Challenges

• Scanners

• Image size: up to 1 GB x 500 slides per day

• No existing infrastructure present for storage of this size at our facility

• Image presentation and software integration

• Logistics

Scanners

• Different manufacturers:

• Speed

• Focusing method

• Acquisition technique

• z-stack acquisition

• File format policy

• Application integration

• 2007: 3D Histech (Zeiss), Aperio, Hamamatsu, Olympus (US: Dmetrix,

BioImagene)

• 2010: Leica, Menarini, Philips, Omnyx, BioImagene

Scanning logistics

• 3 Aperio XT scanners (120 slides per scanner)

• Morgue assistants

• Mark slides after scanning

• 6 Hours per run at 20x magnification (~3 minutes/scan)

Storage – HSM

• HSM = Hierarchical Storage Management• Sun Microsystems (Oracle)• 6 TB available on very fast fibre channel disks• 120 TB available on tape (750 GB each)• 2 Tape drives

• Completely transparent archiving and retrieval (robot)

• Access time from tape: 1 - 3 minutes

Linking systems

• 1D Barcodes• U-DPS: reporting system• LMS: Laboratory Management System• Spectrum: Aperio’s image management solution• Storage system

• Own development: integration layer

Scanner workflow

Image / Data server

Scanning

Database

Storage

Other images (e.g. macro)

pathology users

U-DPS

LIS

HSM

120 TB on tape6 TB fast disks

Order formscanner

Speech recognition

Validation

• Aim: validate diagnostic use of digital slides

• Method: reevaluate diagnosis with same pathologist on scanned

slide after washout period (1 year) for several organs

• Gold standard: original diagnosis using ‘traditional’ microscopy

Preliminary results validation

GI tract (%) Skin (%)

Same 92 94

Similar 7 5

Discrepant 1 1

• GI tract discrepancy

• Different interpretation of abnormality

• Glass slide and WSI contained same information

• Skin discrepancy

• Clinical information not used for interpreting digital slide

Pitfalls of digital archive

• Costs

• Huge storage needs – 40 TeraByte per year (over 57,000 CDs)

• Largest storage in the UMCU

• Logistics of scanning up to 500 slides per day

• Currently scanning almost 24 / 7

• Place of scanning in process

• Speed of image retrieval

• Image compression (JPEG 2000?)

• Backup

Education

• All students view the same “best slide”

• Slide images can be integrated with

• Annotations

• Questions

• Macroscopic images

• Other multimedia

• Most UMCU microscopy practical sessions are digital

• Student satisfaction is high

• Place (small) slide scanners at different labs

• Upload digital slides to UMC Utrecht

• Aurora mScope Clinical

• www.pathoconsult.com

Teleconsultation

Upload new media

Digital slide panel discussions

Image processing research

Image processing applications on virtual slides:• Detecting mitotic figures in breast cancer slides• Use texture features

• Establishing histological grade in breast cancer• Segmentation of individual nuclei (on H&E stained

slides)• Detect points of interest• Use marker-controlled watershed segmentation• Post processing

Conclusions

• Routine scanning is possible and makes sense

• Future of pathology is digital

• Digital pathology is expensive

• Digital pathology is just starting..

Together we are shaping Pathology 2.0

Discussion

• Limitations current system

• Cytology

• Speed

• Magnification (20x / 40x incidental)

• Backup

• Quality Control

Discussion

• Archive heavily used

• Educational use still growing

• Teleconsultation network growing (www.slideconsult.com)

• Need for standards

• DICOM / JPEG 2000

• Images, annotations and reports

• Mixing scanners and integration with other software platforms

(middleware?)

• Image management central in workflow for pathologist?

Questions?!

Huisman et al., Creation of a fully digital pathology slide archive by

high-volume tissue slide scanning, Human Pathology, 2010

May; 41(5): 751-7

[email protected]