Determination of fish gender using fractal analysis of ultrasound images

The gender of cod Gadus morhua can be determined by considering the complexity in their gonadal ultrasonographic appearance. The fractal dimension (D(B)) can be used to describe this feature in images. B-mode gonadal ultrasound images in 32 cod, where gender was known, were collected. Fractal analysis was performed on these images and D(B) was determined using the box counting method. A receiver-operating curve (ROC) was drawn for D(B) as a test for male fish. Using a range of D(B) values, the maximum accuracy for this test was calculated and compared with the accuracy for identifying male fish by subjective analysis alone. The mean (and standard deviation) of the fractal dimension D(B) for male fish was 1.554 (0.073) while for female fish it was 1.468 (0.061); the difference was statistically significant (P = 0.001). The area under the ROC curve was 0.84 indicating the value of fractal analysis in gender determination in cod. Maximum accuracy (0.84) for D(B) as a test for male fish was obtained using the threshold value D(B) = 1.5058 compared with an accuracy of 0.78 for subjective image evaluation. The use of two thresholds, D(B) < 1.4475 (females) and D(B) > 1.5054 (males) gives an 80% certainty in the classification result. Fractal analysis is useful for gender determination in cod. This or a similar form of analysis may have wide application in veterinary imaging as a tool for quantification of complexity in images.