Related Experiment Video
Updated: Oct 29, 2026
Evaporation-reducing Culture Condition Increases the Reproducibility of Multicellular Spheroid Formation in Microtiter Plates
Published on: March 7, 2017
LSTM-ConvNeXt based classification of pre-identified fungal pathogens in culture plate images
Feray Ferda Senol1, Ozlem Aytac1, Mehmet Veysel Gun2
1Elazig Fethi Sekin State of Hospital, Medical Microbiology, Elazig, Turkiye.
Objectives:
To develop a novel lightweight convolutional neural network, named LSTMNeXt, for the automated classification of fungal pathogens from culture plate images and to evaluate its performance on a newly collected fungal colony image dataset.
Materials And Methods:
A fungal colony image dataset comprising 4,080 images from 10 Candida species was collected under standardized imaging conditions. The proposed LSTMNeXt architecture integrates an LSTM-inspired attention mechanism with ConvNeXt- and PoolFormer-based design principles to enhance spatial feature representation while maintaining computational efficiency. The model was pretrained on Stable ImageNet1K and subsequently fine-tuned on the fungal dataset. Performance was assessed using an independent test set with accuracy, precision, recall, and F1-score metrics.
Results:
The proposed LSTMNeXt achieved a test accuracy of 81.37%, with a macro precision of 78.99%, macro recall of 77.75%, and macro F1-score of 77.96%. Regularization and multi-seed analyses demonstrated improved generalization and stable performance across different random initializations.
Conclusions:
LSTMNeXt provides an effective and lightweight solution for fungal pathogen classification from culture plate images. Its LSTM-inspired attention mechanism enables competitive classification performance while maintaining low computational complexity, indicating its potential as a computer-aided tool for microbiological diagnostics.
