Effect of image compression and scaling on automated scoring of immunohistochemical stainings and segmentation of tumor epithelium

Diagnostic Pathology

Johan Lundin⁴  Stig Nordling⁵  Kirsimari Aaltonen³  Heli Nevanlinna²  Carl Blomqvist¹  Caj Haglund⁶  Nina Linder⁴  Mikael Lundin⁴  Juho Konsti⁴ 

[1] Department of Oncology, Helsinki University Central Hospital, Helsinki, Finland;Department of Obstetrics and Gynaecology, Helsinki University Central Hospital, Helsinki, Finland;Department of Clinical Genetics, Helsinki University Central Hospital, Helsinki, Finland;Institute for Molecular Medicine Finland (FIMM), P.O. Box 20, FI-00014 University of Helsinki, Helsinki, Finland;Department of Pathology, University of Helsinki, Helsinki, Finland;Department of General Surgery, Helsinki University Central Hospital, Helsinki, Finland

关键词: Image scaling;    Image compression;    Computer-Assisted;    Image processing;    Texture analysis;    Immunohistochemistry;    Colorectal cancer;    Breast cancer;   

Others  :  808132
DOI  :  10.1186/1746-1596-7-29

 received in 2011-12-22, accepted in 2012-03-21,  发布年份 2012

【 摘 要 】

Background

Digital whole-slide scanning of tissue specimens produces large images demanding increasing storing capacity. To reduce the need of extensive data storage systems image files can be compressed and scaled down. The aim of this article is to study the effect of different levels of image compression and scaling on automated image analysis of immunohistochemical (IHC) stainings and automated tumor segmentation.

Methods

Two tissue microarray (TMA) slides containing 800 samples of breast cancer tissue immunostained against Ki-67 protein and two TMA slides containing 144 samples of colorectal cancer immunostained against EGFR were digitized with a whole-slide scanner. The TMA images were JPEG2000 wavelet compressed with four compression ratios: lossless, and 1:12, 1:25 and 1:50 lossy compression. Each of the compressed breast cancer images was furthermore scaled down either to 1:1, 1:2, 1:4, 1:8, 1:16, 1:32, 1:64 or 1:128. Breast cancer images were analyzed using an algorithm that quantitates the extent of staining in Ki-67 immunostained images, and EGFR immunostained colorectal cancer images were analyzed with an automated tumor segmentation algorithm. The automated tools were validated by comparing the results from losslessly compressed and non-scaled images with results from conventional visual assessments. Percentage agreement and kappa statistics were calculated between results from compressed and scaled images and results from lossless and non-scaled images.

Results

Both of the studied image analysis methods showed good agreement between visual and automated results. In the automated IHC quantification, an agreement of over 98% and a kappa value of over 0.96 was observed between losslessly compressed and non-scaled images and combined compression ratios up to 1:50 and scaling down to 1:8. In automated tumor segmentation, an agreement of over 97% and a kappa value of over 0.93 was observed between losslessly compressed images and compression ratios up to 1:25.

Conclusions

The results of this study suggest that images stored for assessment of the extent of immunohistochemical staining can be compressed and scaled significantly, and images of tumors to be segmented can be compressed without compromising computer-assisted analysis results using studied methods.

Virtual slides

The virtual slide(s) for this article can be found here: http://www.diagnosticpathology.diagnomx.eu/vs/2442925476534995 webcite

【 授权许可】

   
2012 Konsti et al; licensee BioMed Central Ltd.

【 图 表 】

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

【 参考文献 】

• [1]Kayser K, Gortler J, Bogovac M, Bogovac A, Goldmann T, Vollmer E, Kayser G: AI (artificial intelligence) in histopathology--from image analysis to automated diagnosis. Folia Histochem Cytobiol 2009, 47:355-361.
• [2]Konsti J, Lundin J, Jumppanen M, Lundin M, Viitanen A, Isola J: A public-domain image processing tool for automated quantification of fluorescence in situ hybridisation signals. J Clin Pathol 2008, 61:278-282.
• [3]Konsti J, Lundin M, Joensuu H, Lehtimaki T, Sihto H, Holli K, Turpeenniemi-Hujanen T, Kataja V, Sailas L, Isola J, Lundin J: Development and evaluation of a virtual microscopy application for automated assessment of Ki-67 expression in breast cancer. BMC Clin Pathol 2011, 11:3. BioMed Central Full Text
• [4]Mulrane L, Rexhepaj E, Penney S, Callanan JJ, Gallagher WM: Automated image analysis in histopathology: a valuable tool in medical diagnostics. Expert Rev Mol Diagn 2008, 8:707-725.
• [5]Rojo MG, Bueno G, Slodkowska J: Review of imaging solutions for integrated quantitative immunohistochemistry in the Pathology daily practice. Folia Histochem Cytobiol 2009, 47:349-354.
• [6]Gonzalez RC, Woods RE: Digital image processing. 2nd edition. Upper Saddle River: Prentice Hall; 2002.
• [7]Kalinski T, Zwonitzer R, Grabellus F, Sheu SY, Sel S, Hofmann H, Bernarding J, Roessner A: Lossy compression in diagnostic virtual 3-dimensional microscopy-where is the limit? Hum Pathol 2009, 40:998-1005.
• [8]Tengowski MW: Image compression in morphometry studies requiring 21 CFR Part 11 compliance: procedure is key with TIFFs and various JPEG compression strengths. Toxicol Pathol 2004, 32:258-263.
• [9]Lopez C, Lejeune M, Escriva P, Bosch R, Salvado MT, Pions LE, Baugells J, Cugat X, Alvaro R, Jaen J: Effects of image compression on automatic count of immunohistochemically stained nuclei in digital images. J Am Med Inform Assoc 2008, 15:794-798.
• [10]Lopez C, Martinez JJ, Lejeune M, Escriva P, Salvado MT, Pons LE, Alvaro T, Baucells J, Garcia-Rojo M, Cugat X, Bosch R: Roundness variation in JPEG images affects the automated process of nuclear immunohistochemical quantification: correction with a linear regression model. Histochem Cell Biol 2009, 132:469-477.
• [11]Lejeune M, Lopez C, Bosch R, Korzynska A, Salvado MT, Garcia-Rojo M, Neuman U, Witkowski L, Baucells J, Jaen J: JPEG2000 for automated quantification of immunohistochemically stained cell nuclei: a comparative study with standard JPEG format. Virchows Arch 2011, 458:237-245.
• [12]Barker NJ, Zahurak M, Olson JL, Nadasdy T, Racusen LC, Hruban RH: Digital imaging of black and white photomicrographs--Impact of file size. Am J Surg Pathol 1998, 22:1411-1416.
• [13]Ojala T, Pietikainen M, Maenpaa T: Gray scale and rotation invariant texture classification with local binary patterns. Computer Vision - Eccv 2000, Pt I, Proceedings 2000, 1842:404-420.
• [14]Martens G, Poppe C, Lambert P, Van de Walle R: Noise- and compression-robust biological features for texture classification. Vis Comput 2010, 26:915-922.
• [15]Ahlin C, Aaltonen K, Amini RM, Nevanlinna H, Fjallskog ML, Blomqvist C: Ki67 and cyclin A as prognostic factors in early breast cancer. What are the optimal cut-off values? Histopathology 2007, 51:491-498.
• [16]Linder N, Konsti J, Turkki R, Rahtu E, Lundin M, Nordling S, Ahonen T, Pietikainen M, Lundin J: Identification of tumor epithelium and stroma in tissue microarrays using texture analysis. Diagn Pathol 2012, 7:22. BioMed Central Full Text
• [17]Entwistle A: Formats of image data files that can be used in routine digital light micrography. Part two. Biotech Histochem 2003, 78:91-99.