A Deep Learning Approach to MicroorganismImage Analysis Using MATLAB

In this paper, we propose a new deep-learningbased framework for the automatic analysis of microorganism images in MATLAB. In fact, this is an effort to improve the accuracy and efficiency of the microorganism classification method using the Convolutional Neural Network (CNN) approach. Our dataset was very inhomogeneous, containing many different images of microorganisms, so we first normalized with respect to image quality and size. We then developed, balancing between model complexity and computational efficiency, training our custom-made CNN model. We quantitatively evaluate our model by using rigorous experiments with standard metrics of accuracy, precision, recall, and F1 score. The experimental results of our model could provide following accuracies: the model has an accuracy of 83.78%, precision of 83.88%, recall of 86.72%, and, hence, an F1 score of 84. Such metrics point to the high reliability of the model in the process of microorganism identification, possibly with a slightly higher level of recall pointing to fewer false-negative cases in the classification of the biological entity. Further analysis suggests that in addition to becoming a strong solution for the classification of microorganisms, our model also becomes a strong basis for the following such research. Future work could integrate larger datasets, transfer learning, and even deploy the model in a real clinical setting. Our work shows the possibilities of applying deep learning with MATLAB as a powerful stronghold for the development of microorganism image analysis, which is critical in many applications aimed at the health care, environmental monitoring, and biological research fields.