BMC microbiology · 2026
Comparative evaluation of deep learning architectures for microbial colony classification in microbiological imaging.
СтатьяСравнительное исследованиеОценочное исследованиеНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Ahmed Mohamed Abdelmoaty, Alenazy Rawaf, AbdelMoety Ahmed, Kamel Hend A, Shafik Shokri M
Аннотация
BACKGROUND: Visual interpretation of microbial colonies on agar plates is an essential task for experienced microbiologists. In high-throughput workflows and large-scale screening, automated image-based methods can provide decision-support capabilities in standardized laboratory imaging setups and consistency. Deep Learning (DL) can support automated identification, but comparative evidence across widely used Convolutional Neural Networks (CNNs) on the same dataset remains limited.
METHODS: This study performed a head-to-head comparison of six CNN architectures-AlexNet, SqueezeNet 1.1, ResNet18, ShuffleNetV2, EfficientNetB0, and MobileNetV2-for microbial colony classification. Using the public AGAR dataset, we processed 5,269 plate images and used the provided annotations to extract 86,045 colony crops from 4,424 countable plates representing five species (Bacillus subtilis (B. subtilis), Escherichia coli (E. coli), Pseudomonas aeruginosa (P. aeruginosa), Staphylococcus aureus (S. aureus), and Candida albicans (C. albicans)). Crops were resized to 224 × 224 and augmented (random rotation, horizontal flip, and color jitter). All models were trained under a unified protocol (Adam, learning rate [Formula: see text], batch size 32, early stopping) with a split (70% training, 15% validation, 15% testing). Performance was assessed using accuracy, precision, recall, and F1-score.
RESULTS: MobileNetV2 achieved the highest test accuracy (99.40%), followed closely by EfficientNetB0 (99.38%), ShuffleNetV2 (99.19%), and ResNet18 (99.16%), while SqueezeNet 1.1 also showed strong performance (99.09%) and AlexNet yielded the lowest accuracy (98.74%). These findings indicate that, although modern lightweight architectures performed best, ResNet18 and SqueezeNet 1.1 remained highly competitive under the unified evaluation protocol.
CONCLUSION: Modern CNN architectures, especially MobileNetV2 and EfficientNetB0, offer accurate microbial colony classification on standardized colony crops. They show promise for improving automated microbiological image analysis. This approach could facilitate integration into medical equipment, reduce manual labor, expedite diagnosis, and save costs.
METHODS: This study performed a head-to-head comparison of six CNN architectures-AlexNet, SqueezeNet 1.1, ResNet18, ShuffleNetV2, EfficientNetB0, and MobileNetV2-for microbial colony classification. Using the public AGAR dataset, we processed 5,269 plate images and used the provided annotations to extract 86,045 colony crops from 4,424 countable plates representing five species (Bacillus subtilis (B. subtilis), Escherichia coli (E. coli), Pseudomonas aeruginosa (P. aeruginosa), Staphylococcus aureus (S. aureus), and Candida albicans (C. albicans)). Crops were resized to 224 × 224 and augmented (random rotation, horizontal flip, and color jitter). All models were trained under a unified protocol (Adam, learning rate [Formula: see text], batch size 32, early stopping) with a split (70% training, 15% validation, 15% testing). Performance was assessed using accuracy, precision, recall, and F1-score.
RESULTS: MobileNetV2 achieved the highest test accuracy (99.40%), followed closely by EfficientNetB0 (99.38%), ShuffleNetV2 (99.19%), and ResNet18 (99.16%), while SqueezeNet 1.1 also showed strong performance (99.09%) and AlexNet yielded the lowest accuracy (98.74%). These findings indicate that, although modern lightweight architectures performed best, ResNet18 and SqueezeNet 1.1 remained highly competitive under the unified evaluation protocol.
CONCLUSION: Modern CNN architectures, especially MobileNetV2 and EfficientNetB0, offer accurate microbial colony classification on standardized colony crops. They show promise for improving automated microbiological image analysis. This approach could facilitate integration into medical equipment, reduce manual labor, expedite diagnosis, and save costs.