Virtual Microscopy
Design and Implementation of Novel Software Applications for Diagnostic Pathology
Acta Universitatis Tamperensis No. 1720
By Tuominen Vilppu
July 2012
Tampere University Press
ISBN: 9789514487736
186 pages
$87.50 Paper original
Accurate histopathological diagnosis is an essential part of clinical cancer patient treatment, and is used, for example, to determine the patient’s eligibility for surgical and adjuvant therapies, such as chemotherapy. Pathologists reach the diagnosis by inspecting stained tumor section slides with a light microscope or, increasingly, by digitizing the specimens and inspecting them with a computer display. The digitization of an entire microscope specimen at a diagnostically adequate resolution and the subsequent data processing are collectively referred to as virtual microscopy. Similarly, the digitized specimen slides are referred to as virtual slides. Although currently available technology enables routine usage of virtual microscopy, the amount of data generated with high-throughput virtual slide scanning is enormous–up to hundreds of gigabytes of uncompressed data per slide. Processing the data requires specialized information technology methods, which differ considerably from other medical imaging disciplines. Automated specimen scanning, processing, image analysis, archival, linkage to clinical information systems, and distribution to the end-users all present their own unique challenges to the software and hardware development. The aim of the present thesis study is to identify these problems and solve them by designing and implementing an open and standards-based software platform, which will facilitate the large-scale usage of virtual microscopy in clinical pathology, research, and education.
The sample material of the study consisted primarily of histological tumor section slides and secondarily of radiological imagery. The slides were digitized using various commercial and in-house scanning systems, for which an automated slide acquisition controller (DirObserver) and a stitching software application (LargeMontage) were developed. The suitability of JPEG2000 image compression standard for virtual microscopy, its compression efficiency, performance, and the optimal code-stream parameterization were studied, and based on these results, two software packages were designed and implemented. The first package allows the utilization of JPEG2000 in virtual microscopy and consists of three applications: a virtual slide viewer (JVSview), a slide server (JVSserv), and a slide converter (JVScomp). The second package provides proof-of-concept software for linking virtual slides with clinical information systems and Picture Archiving and Communication System (PACS) -based image databases, which follow the Digital Imaging and Communications in Medicine (DICOM) standard in image data exchange. The package consists of three applications: a DICOM PACS client (JVSdicom Workstation), a DICOM PACS server (JVSdicom Server), and a DICOM image converter (JVSdicom Compressor). For the automated image analysis of immunohistochemical (IHC) samples stained for the breast cancer biomarkers estrogen receptor (ER), progesterone receptor (PR), Ki-67, and Human Epidermal growth factor Receptor 2 (HER2), two web-based image analysis applications were developed (ImmunoRatio and ImmunoMembrane), which were calibrated to match the visual assessment of expert pathologists. The software development for the present study was done using various programming languages, libraries, frameworks, and development environments. All the executable binaries, web applications, and/or software source code have been released for free and public use on our research group website http://jvsmicroscope.uta.fi/.
The virtual microscopy software platform we have developed is currently being used in several academic and clinical institutions throughout Finland, and there has been significant interest from abroad as well. We have shown that JPEG2000 is a viable solution as the universal virtual slide format, readily linkable with clinical information systems. By using the image analysis software we have described, routine clinical diagnostics of ER, PR, Ki-67, and HER2 IHC can be made in shorter overall analysis time, while improving the reproducibility and repeatability of the analysis. We anticipate that virtual microscopy will continue to gain momentum in the clinical pathology diagnostics, research, and education in the near future.