Document not found! Please try again

Temporal and spatial variability in Baltic sprat batch ...

2 downloads 0 Views 727KB Size Report
The importance of fecundity for studying fish reproductive ecology has led to many research efforts to provide easier, faster and low-cost methods. Fecundity is ...
New automatic software tool to estimate fish fecundity based on image analysis Rosario Dominguez-Petit1, José Manuel Pintor Freire2, Sonia Rábade1, Mariña Fabeiro1, Ángel Dacal-Nieto2, Pilar Carrión2, Encarnación González-Rufino2, Eva Cernadas3, Manuel Fernández-Delgado3, Dolores Dominguez-Vázquez1, Fran Saborido-Rey1, Arno Formella2 1

Department of Fisheries Ecology, Institute of Marine Research - CSIC, Vigo, Spain. 2 Department of Computer Science, University of Vigo, Spain. 3 Department of Electronics and Computer Science, University of Santiago de Compostela, Spain.

Introduction The importance of fecundity for studying fish reproductive ecology has led to many research efforts to provide easier, faster and low-cost methods. Fecundity is estimated counting and measuring the matured oocytes in histological images of fish ovaries. Govocitos is a multi-platform software to automate count, classification and measurement of oocytes, and estimates fecundity from histological images based on stereological principles, a functionality which is not provided by other available software tools.

Methodological and software specifications The image acquisition process uses standard histological procedures to section the ovaries. The sections are stained with haematoxylin–eosin and captured with LEICA ® equipment with a spatial resolution of 1.09 μm. The exposure time and color balance are set automatically. Govocitos 4.0 is written in C++ to analyze histological images including the following modules: 1) a user friendly interface to manually draw the outline and classify oocytes as well as to automatically measure them; 2) supervised and unsupervised automatic detection and classification and 3) automatic fecundity estimation. The information required and calculated by Govocitos is supported by either local or web-based databases and XML files.

The unsupervised edge detection algorithm has two steps: the image is processed by a MultiScalar Canny filter, followed by an edge analysis step to determine which edges are candidates to represent a true oocyte. In the supervised algorithm, a point inside the true oocyte must be provided to help the edge analysis algorithm. Oocytes are automatically classified by Support Vector Machine classifiers that use texture-based methods. When new specie images are processed for the first time, Govocitos has to be trained. In the present work it has been trained for European hake (Merluccius merluccius, MZ), pouting (Trisopterus luscus, TL) and lane snapper (Lutjanus synagris, LS).

Fecundity was estimated according to Emerson’s formula: F = Ov ×

k

β

×

N a3 vi

Where β is a shape coefficient, K is a size distribution coefficient, Na is the number of profiles in each image, vi is the volume fraction of each oocyte within the image and Ov is the ovary volume.

Results

Table 1 shows results of automatic detection and classification for each species and algorithm. Automatic detection takes 20 seconds using a Intel® Core™2 Quad Processor. Classification

Unsupervised detection

Supervised detection

Maturity stage

N/NN

MZ

75.0%

90.0%

87.5%

81.0%

TL

75.0%

90.0%

80.2%

74.8%

LS

75.0%

90.0%

91.0%

82.1%

Table 1. Percentage of correctly detected and classified oocytes for each studied species. N/NN: nucleus/ nonnucleus.

Figure 1 shows the percentage of correctly detected cells depending on the tolerance level (T) for both supervised and unsupervised algorithms as well as the noise factor for the unsupervised algorithm. Supervised method will be implemented in Govocitos in the near future. Tolerance is the overlap area between the automatic software detected and manually drawn oocyte (0.5

Suggest Documents