Separation enhanced nucleus detection on propidium iodide stained digital slides V.Z. Jonas * , M. Kozlovszky * , B. Molnar** * Obuda University/John von Neumann Faculty of Informatics, Budapest, Hungary **Second Department of Internal Medicine, Semmelweis University, Budapest, Hungary. Abstract — Cancer research and diagnostics is an important frontier to apply the power of computers. Image processing gains more and more territory in pathology, our current study aims to apply image cytometry to this area. One of the most fundamental methods is ploidy analysis that aims to measure the pace of proliferation in the tissue. This is proven to be a good marker of tumor presence and aggressivity. Ploidy analysis traditionally is a flow cytometry area. The samples used in our study are the same as those used in a flow cytometer: a droplet of the suspension containing stained nuclei caught on a glass slide, covered and scanned. This article aims to show our progress in nucleus detection on digitized samples of propidium iodide stained cell nuclei suspensions. A previous article covered the assay method constructed and the initial segmentation technique used for nuclei detection, and identified its shortcomings based on the results. This paper presents a step forward in segmentation accuracy in our project. I. INTRODUCTION As part of a project [1] aiming to prove the advantages of Image Cytometry (ICM) in ploidy analysis we try to formulate a segmentation algorithm as precise as possible. The reference to compare our algorithm to is usually in the head of experts working, often overtime to diagnose patients. We are working on a validation method for algorithms on this area of expertise that demands the least possible amount of their time [2]. Also in parallel we use this method to measure our progress in perfecting the segmentation algorithm to ensure to produce the most exact measurements possible. An article published earlier [3] discusses the validation method and the segmentation method. This article shows how the segmentation algorithm has been changed, and how that change influences the measurement results, confirming it as an improvement. The segmentation algorithm processed 8 bit one channel fluorescent image containing the light emitted by the propidium iodide (PI) dye, as visible on Figure 1. Figure 1. Propidium iodide stained nuclei on a digital slide. . PI binds to nucleotic acids thus making the DNA visible to this imaging technique. Image brightness is proportional to the DNA content. II. MATERIALS AND METHODS The assay is based on 17 specimens of healthy human blood sample containing only Propidium Iodide stained cell nuclei. These were digitized using a glass slide scanner produced by 3DHistech Ltd. (Pannoramic Scan, fluorescent setup, 5MP sCMOS camera, 40x (Zeiss) objective, and a LED-based lighting method to excite the fluorochrome.) The resulting resolution of the samples were 0.1625µm/per pixel (jpeg compressed). Three of the twenty scanned digital slides from of the specimen were not sufficient in quality. (The scanning method and parameters are also being formed.) These three samples were excluded from the further validation process, re- scanning not being an option, because of the dye bleaching effect. III. DISCUSSION This article is strongly founded on our previously published results. We worked with a pathology expert to help us locate and mark all nuclei on each the sample in a 1mm 2 area, used our segmentation algorithm to detect the same nuclei. The concluding step was to compare the results of the algorithm and the expert. To achieve this we defined four possible classes, and labelled every object of the segmentation result for all the 17 samples. Class Nucleus is the nuclei marked by the expert (control or reference); all other classes are the classes the detection results are divided to. The resulting classes were: - Match: one segmentation result for one expert- marked nucleus - FP: false positive (no marker by expert, algorithm finds object) - FN: false negative (expert-marked nucleus present, no segmentation result) - Other: separation problem (multiple expert-markers within a segmentation result) – 157 – INES 2014 • IEEE 18th International Conference on Intelligent Engineering Systems • July 3-5, 2014, Tihany, Hungary 978-1-4799-4615-0/14/$31.00 ©2014 IEEE