The participants are required to design and implement a pattern recognition system able to classify a whole HEp-2 specimen image. Each specimen image belongs to one of the following pattern classes: homogeneous, speckled, nucleolar, centromere, golgi, nuclear membrane and mitotic spindle.
Classes of staining patterns
- homogeneous : a uniform diffuse fluorescence covering the entire nucleoplasm sometimes accentuated in the nuclear periphery
: these patterns have two sub-categories:
- coarse speckled : densely distributed, variously sized speckles, generally associated with larger speckles, throughout nucleoplasm of interphase cells; nucleoli are negative
- fine speckled : fine speckled staining in a uniform distribution, sometimes very dense so that an al- most homogeneous pattern is attained; nucleoli may be positive or negative
- nucleolar : brightly clustered large granules corresponding to decoration of the fibrillar centers of the nucleoli as well as the coiled bodies
- centromere : rather uniform discrete speckles located throughout the entire nucleus
- golgi : staining of a polar organelle adjacent to and paratly surrounding the nucleus, composed of irregular large granules. Nuclei and nucleoli are negative. Diffuse staining of the cytoplasm of dividing cells sometimes with accentuation around chromosomal material
- nuclear membrane : a smooth homogeneous ring-like fluorescence of the nuclear membrane in interphase cells
- mitotic spindle : staining only of the triangular or “banana-shaped” pole area of the mitotic spindle in the metaphase cells
For each specimen to be classified, the system receives as input four images taken at different locations of the specimen and the respective segmentation masks together with the intensity level information (positive or intermediate). The segmentation mask contains all the object present into the specimen, thus not only the cells in the interphase but also cells in mitosis and artifacts (air bubbles, etc).
The dataset was acquired in 2013 at Sullivan Nicolaides Pathology Laboratory, Australia. It was collected from 1001 patient sera with positive ANA test. Each patient sera were diluted to 1:80 and the specimen was photographed using a monochrome camera fitted on a microscope. Each specimen was photographed at four different locations rendering each specimen has four images. The dataset has seven pattern classes: homogeneous, speckled, nucleolar, centromere, nuclear membrane, golgi and mitotic spindle. The first four classes represent common ANA patterns whilst the rest three classes are less common.
The dataset is divided into training and test set as following: approximately ¼ sera in the training set and the remaining in the test set. All images are in monochromatic uncompressed format with resolution 1388 x 1040 pixels together with their corresponding cell mask which was obtained automatically.
The labelling process has involved at least two scientists who read each patient specimen under a microscope. A third expert’s opinion has been sought to adjudicate any discrepancy between the two opinions. Each slide image is also provided with the corresponding class (one among the patterns defined above). Furthermore, all the labels were validated by using secondary tests such as ENA and anti-ds-DNA in order to confirm the presence and/absence of specific patterns.
Each specimen image contained in the database is annotated with the following information:
- Staining pattern of the specimen
- Intensity of the specimen
- Mask of the specimen
The application obtaining the highest value of the mean class accuracy in specimen classification over the test set will be declared as the winners. The proclamation of the winners will be made during the contest session at ICPR 2014. The contest session will be held as one of the sessions of the I3A Workshop.