CHARACTERIZATION AND AUTOMATIC COUNTING OF F.I.S.H. SIGNALS IN 3-D TISSUE IMAGES

• Published in: Image analysis & stereology Vol. 20; no. 1; pp. 41 - 52
• Main Authors: Adiga, Umesh PS, Knight, Samantha JL, Chaudhuri, BB
• Format: Journal Article
• Language: English
• Published: Slovenian Society for Stereology and Quantitative Image Analysis 03.05.2011
• Subjects: characterization; counting; features; FISH signal
• ISSN: 1580-3139; 1854-5165
• DOI: 10.5566/ias.v20.p41-52

The evaluation of malignancy-related features often helps to determine the prognoses for patients with carcinomas. One technique, which is becoming increasingly important for assessing such prognostic features is that of Fluorescence in situ Hybridization (FISH). By counting the number of FISH signals in a stack of 2- D images of a tumor (which together constitute the 3-D image volume), it is possible to determine whether there has been any loss or gain of the target DNA sequences and thereby evaluate the stage of the disease. However, visual counting of the FISH signals in this way is a tedious, fatiguing and time-consuming task. Therefore, we have developed an automated system for the quantitative evaluation of FISH signals. We present and discuss the implementation of an image processing module that segments, characterizes and counts the FISH signals in 3-D images of thick prostate tumor tissue specimens. Possible errors in the automatic counting of signals are listed and ways to circumvent these errors are described. We define a feature vector for a FISH signal and describe how we have used the weighted feature vector to segment specific signals from noise artifacts. In addition, we present a method, which allows overlapping FISH signals to be distinguished by fitting a local Gaussian model around the intensity profile and studying the feature vector of each model. Our complete image processing module overcomes the problems of manual counting of FISH signals in 3-D images of tumor specimens, thereby providing improved diagnostic and prognostic capability in qualitative diagnostic pathology.