Leveraging Smartphone Imaging and Deep Transfer Learning for Bacterial Colony Classification: From Uncontrolled to Controlled Settings

The classification of bacterial colony types is crucial for microbiology and medical diagnostics because it provides essential information for pathogen identification and treatment planning. The traditional classification methods are resource-intensive approaches and require specialized equipment and expertise. In this study, we explored the integration of smartphone imaging and transfer learning techniques for bacterial colony classification purposes. We expanded an existing dataset of Escherichia coli (E. coli) and Klebsiella pneumoniae (K. pneumoniae) bacteria grown on MacConkey (MAC) agar by incorporating Pseudomonas aeruginosa (P. aeruginosa) on the same medium, as well as Staphylococcus aureus (S. aureus), P. aeruginosa, K. pneumoniae, and E. coli on brain–heart infusion (BHI) agar. The total numbers of images in the MAC agar and BHI agar datasets were 9819 and 12348, respectively. Utilizing state-of-the-art convolutional neural networks (CNNs) such as SqueezeNet1.0, MobileNetV3 Small, ShuffleNetV2 X1.5, EfficientNetB0, DenseNet121, and ResNet18, we evaluated the classification performance achieved across uncontrolled and controlled imaging environments. We also compared the results produced for the same bacterial type on different media (MAC agar and BHI agar). The results showed that the lightweight models, i.e., EfficientNetB0 and ShuffleNetV2 X1.5, consistently performed well across different datasets, reaching 100% accuracy under controlled conditions. We believe that this work provides a foundation for achieving cost-effective bacterial colony classification on smartphones using transfer learning techniques, ultimately enabling practical implementations. Future work will focus on enhancing the explainability of the model and evaluating its deployment feasibility on smartphone devices.