Towards Fully Automatic Acquisition of Multimodal Cytopathological Microscopy Images with Autofocus and Scene Matching

Summary

Objectives: To increase the chance for a cure, cancer must be detected as early as possible. This can be achieved with cytopathological diagnostic methods. For a further increase of the diagnostic accuracy of these methods we introduced the multimodal cell analysis, viz, cells on the slide have to be relocalized to enable successive analysis of identical cells in different stains. For practical reasons the relocalization step must be automated.

Methods: For a fully automatic acquisition of successive cell images we use a passive autofocus that is adaptive to the material, i.e., to the cells, followed by a comparison of the scenes, i.e., the cell constellation, of two such obtained images from different stains. In case that no sub-scene match can be found the search is extended to the surrounding area. A set of 1 556 scenes from seven specimens have been subject to our algorithm. The automatically relocalized and acquired images from a second stain have been manually compared to the images from a first stain.

Results: An overall relocalization rate of 85.4% is achieved. 14.3% of the images could not be relocalized and are lost for the following diagnostic process, while the critical case of erroneously matched images was observed in only 0.3% of cases.

Conclusions: We could show that it is possible to automatically acquire images of successive stains of identical cells on cytopathological specimens. The method presented achieves acceptable relocalization rates. Wrong image acquisitions are very rare and can mostly be ascribed to images with single cells, i.e., without scene information.

Keywords

Cytopathology - multimodal microscopy - autofocus - scene detection - automation

• References

• 1 Carpi A, Ferrari E, Toni MG, Sagripanti A, Nico-lini A, Di Coscio G. Needle aspiration techniques in preoperative selection of patients with thyroid nodules: along-termstudy. JClin Oncol 1996; 14 (05) 1704-1712.
• 2 Remmerbach TW, Weidenbach H, Pomjanski N, Knops K, Mathes S, Hemprich A, Böcking A. Cytologic and DNA-cytometric early diagnosis of oral cancer. Anal Cell Pathol 2001; 22 (04) 211-221.
• 3 Böcking A, Stockhausen J, Meyer-Ebrecht D. Towards a single cell cancer diagnosis. Multimodal and monocellular measurements of markers and morphology (5M). Cell Oncol 2004; 26 1-2 73-79.
• 4 Viola P, Wells WM III. Alignment by maximization of mutual information. Int Conf Computer Vision. 1995: 16-23.
• 5 Collignon A, Vandermeulen A, Suetens P, Marchal G. 3D multi-modality medical image registration based on information theory. Computational Imaging and Vision 3. Kluwer Academic. 1995: 263-274.
• 6 Würflinger T, Stockhausen J, Meyer-Ebrecht D, Böcking A. Robust Automatic Coregistration, Segmentation and Classification of Cell Nuclei in Multimodal Cytopathological Microscopic Images. Computerized Medical Imaging and Graphics 2004; 28 1-2 87-98.
• 7 Modersitzki J. Numerical Methods for Image Registration. Oxford University Press; 2004
• 8 Tenenbaum JM. Accommodation in computer vision. PhD thesis.. Stanford University; 1970
• 9 Nayar SK, Nakagawa Y. Shape from focus. IEEE Trans Pattern Analysis and Machine Intelligence 1994; 16 (08) 824-831.
• 10 Jarvis RA. Focus optimization criteria for computer image processing. Microscope 1976; 24 (02) 163-180.
• 11 Firestone L, Cook K, Culp K, Talsania N, Preston K. Comparison of Autofocus Methods for Automated Microscopy. Cytometry 1991; 12: 195-206.
• 12 Kang-Sun C, Jun-Suk L, Sung-Jae K. New auto-focusing technique using the frequency selective weighted median filter for video cameras. IEEE Trans Consumer Electronics 1999; 45 (03) 820-827.
• 13 Otsu N. A threshold selection method from gray level histograms. IEEE Trans Systems, Man and Cybernetics 1979; 09: 62-66.
• 14 Dougherty ER. An introduction to morphological image processing. SPIE. 1992