Classification of the growth level of fungal colonies in solid medium: a machine learning approach
The measurement of colony growth in a solid medium is commonly applied in phytopathology. The measurement procedures usually involve the visual identification and manual measurement of colonies in Petri dishes. Recently, some measurement techniques have been developed based on the segmentation of co...
Saved in:
Published in | Expert systems with applications Vol. 232; p. 120872 |
---|---|
Main Authors | , , , , |
Format | Journal Article |
Language | English |
Published |
Elsevier Ltd
01.12.2023
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The measurement of colony growth in a solid medium is commonly applied in phytopathology. The measurement procedures usually involve the visual identification and manual measurement of colonies in Petri dishes. Recently, some measurement techniques have been developed based on the segmentation of colony images by using digital image analysis and machine learning models. These approaches have in common: a highly controlled environment where the images were obtained; and a segmented image as output. In addition, few machine learning studies explore the importance of image features in the classification process and test a minimal range of algorithms. Thus, in this study, we propose a classification method of fungal growth based on machine learning, which performs this task using images obtained without any control of the luminosity conditions. An image dataset was created with a set of 537 images of Petri dishes incubated with Botrytis cinerea and obtained in the phytopathology laboratory of the UTFPR. The images were preprocessed, and a set of 97 feature descriptors extracted from them generated five different datasets. Experiments were carried out with 9 different classification algorithms trained through repetitive cross-validation resampling. Induced models were compared in terms of the Balanced Accuracy per Class and F-Score metrics, with statistical significance verified through the Kruskal–Wallis test. From all trained models, 15 showed both balanced accuracy and F-score values above 0.82, and the top 12 showed no difference at 5% of significance by the statistical test. Among the evaluated features, the color channels were the most relevant, especially the standard deviation of the intensity channel.
•The manual measurement of fungal growth in the solid medium is a laborious routine.•Growth of fungal colonies can be predicted through image classification analysis.•This analysis involved an extensive selection of image features.•Application and selection of machine learning algorithms.•Our work provides a modular and extensible framework to approach this problem. |
---|---|
ISSN: | 0957-4174 1873-6793 |
DOI: | 10.1016/j.eswa.2023.120872 |