Design and evaluation of a novel digital pathology workstation for clinical use Darren Treanor BSc (Computing) MB BCh PhD FRCPath Leeds Teaching Hospitals Trust & University of Leeds 12 th European Congress on Digital Pathology, Paris, June 2014 darrentreanor@nhs.net @LeedsPathology
Context Leeds Teaching Hospitals NHS Trust 35 consultant pathologists 30 trainee pathologists ~ 250,000 H&E slides/ year in histopathology Scanning since 2003 4 Aperio scanners >180,000 slides 160TB of image data This is independent research carried out by us at the University of Leeds/ Leeds Teaching Hospitals NHS Trust funded by the UK National Institute for Health Research (NIHR) Conflict of interest statement: None to declare
All at Website: 10,000 + virtual slides, slide library, e-learning, QA materials, papers, videos and more Powerwall: 48 megapixels, size 3.5m x 1.5m, 2 on site at Hospital, 2 in University
Background: There s something wrong with digital pathology... Poor acceptability Poor efficiency Majority of pathologists feel virtual slides are not suitable for clinical diagnosis 60% less efficient than the microscope The virtual slide and conventional microscope a direct comparison of their diagnostic efficiency. Annual Meeting of the Pathological Society of Great Britain and Ireland. 2007, Glasgow.
Why are virtual slide viewers less efficient than the microscope? Display resolution Software performance (moving pixels) Software design GUI design adopted from photo editing software Lack of training and experience Other factors Lower image quality than microscope, at both image capture and display stages
Why are virtual slides less efficient? The effect of display resolution 17 xga monitor shows 14% of the viewable area of the microscope field at 40x
Project overview To make a digital microscope as fast as, or faster than, a conventional microscope Study work practices and workflow Develop a digital microscope Evaluate the system with pathologists
What we did Studied work in pathology 1. Quantitative Workload metrics Quantification of range of task types performed with the microscope 2. Qualitative Observations, interviews Videos of work in the office, using the microscope 3. Experimental Evaluation of workstation efficiency, utility Evaluation of specific UI questions Randell R, Ruddle RA, Thomas R, Treanor D. Diagnosis at the microscope: a workplace study of histopathology. Cognition, Technology & Work 2011 Aug.
Working in the office: time breakdown 4% 10% 2% 2% 1% 1% 3% 23% 54% Viewing slide Dictation Other Computer Notes Annotating Book Paperwork Filling forms Randell R, Ruddle RA, Quirke P, Thomas RG, Treanor D. Working at the microscope: analysis of the activities involved in diagnostic pathology. Histopathology 2011
Tracking at the microscope a simple skin specimen Pathologists make hundreds of movements when viewing a slide Mean fixation time 1.3 2.0 seconds Follow complex paths The microscope is transparent to them they focus on the image, the pathology
Project overview To make a digital microscope as fast as, or faster than, a conventional microscope Study work practices and workflow Develop a digital microscope Evaluate the system with pathologists
Design, discussion, prototyping, more discussion
Leeds virtual microscope v2: Final design Fast pixel moving engine Medical grade monitors Barco Coronis 6MP plus Nio 3MP Total 9 megapixels equivalent visual FOV and resolution as a microscope Novel case view Viewing multiple slides (26% of our slide workload at Leeds) Increased reference to metadata ( doing 2 things at once ) Generating complex reports Microscope Workstation
Workstation version 2 design: Screenshot
Slides with metadata displayed in 2D grid
Leeds virtual microscope v2.0 video
Project overview To make a digital microscope as fast as, or faster than, a conventional microscope Study work practices and workflow Develop a digital microscope Evaluate the system with pathologists
Results: Similar time to complete the task on LVM Device Mean (%) Median (%) Exposure Leeds virtual microscope 100.8 96.5 30 minutes Microscope 99.3 112 10 years Normalised time to complete task Error bars show 95% confidence interval (CI)
Higher percentage of time spent viewing slides on LVM p <.05 Mean % time spent viewing slides Error bars show 95% confidence interval (CI)
Tracking shows pathologist returning to slides Suggests that confirming diagnosis, rather than relying on memory - or less confident of decision?
And more acceptable to pathologists!
Summary 5 year effort User centred design Multidisciplinary team Created Powerwall Desktop LVM Equivalent time to the microscope More time spent looking at the slides Acceptable to users
Acknowledgements Virtual Pathology at University of Leeds Phil Quirke David Turner, Martin Waterhouse, Mike Hale, Alex Wright, Fraser Lewis, Hannah Dee, Gordon Hutchins Nick Roberts, Richard Quirke, Mike Shires, Doreen Crellin, Nick Roberts (technical work) Image analysis & computer vision at University of Leeds Alex Wright & Andrew Bennett &Yanong Zhu, Pathology and Tumour Biology Group, Leeds Institute of Molecular Medicine, University of Leeds Ladislav Gubic, Jim Swainston, James Bridges, BSc students, School of Computing, University of Leeds Andy Bulpitt, Derek Magee, Yi Song, Yu Zhou, School of Computing, University of Leeds Visualisation & human computer interaction at University of Leeds Roy Ruddle, Rhys Thomas, John Hodrien School of Computing Rebecca Randell, Leeds Institute of Molecular Medicine Collaborators/ supporters Claudia Mello-Thoms, University of Pittsburgh David Brettle, Leeds Teaching Hospitals Trust Barco Funding Pathological Society of Great Britain and Ireland Department of Health Yorkshire Cancer Research National Institute for Health Research