Direct identification of fungi using image analysis

Filamentous fungi have often been characterized, classified or identified with a major emphasis on macromorphological characters, i.e. the size, texture and color of fungal colonies grown on one or more identification media. This approach has been rejcted by several taxonomists because of the subjectivity in the visual evaluation and quantification (if any)of such characters and the apparent large variability of the features.

We present an image analysis approach for objective identification and classification of fungi. The approach is exemplified by several isolates of nine different species of the genus Penicillium, known to be very difficult to identify correctly. The fungi were incubated on YES and CYA for one week at 25 C (3 point inoculation) in 9 cm Petri dishes.

The cultures are placed under a camera where a digital image of the front of the colonies is acquired under optimal illumination conditions and with an accurately calibrated acquisition system. A PC-based system has been developed. After acquisition and image correction it automatically determines the size and position of the colonies in the digitized image.

Having localized the colonies it is possible to extract a number of different features to distinguish between different species. The results are surprisingly impressive. Classification of 151 colonies from the YES media and 194 colonies from the CYA media was done using a quadratic classifier based on a normal distribution.

In both cases the result was 99% correctly classified colonies. The key to this success is the factor of objectiveness obtained through the application and selection of species specific features that are embedded in the color, texture and size of the colonies. Automatic identification of fungi could be a future perspective, but more criteria or media should probably be included if a larger number of species need to be identified.