Answers are grounded in this paper and cite their sources.
The need for precise classification of fungi from microscopic images plays a critical role in diagnosing and treating patients since it requires timeliness and effectiveness. Nevertheless, manual examination is time-consuming and difficult due to the similar visual features of numerous fungi. Therefore, this work proposes a new approach to explainable deep learning for fungal image classification and report generation. The developed framework consists of four pre-trained convolutional neural network models—DenseNet121, ResNet18, VGG16, and InceptionV3—which are used to classify fungal microscopic images into five classes. To make the process more transparent, methods such as Grad-CAM, SHAP, LIME, and Canny edge detection are used to identify the areas of images that contribute to the prediction. At the same time, the developed approach uses the large language model LLaMA3-70B through LangChain and provides automatic generation of the structured diagnostic report, including fungal identification, confidence scores, and clinical observations. The whole pipeline is implemented using the Streamlit interface to predict and generate a PDF report of the classification. The experiments have been performed using publicly available data containing 5000 images of fungi with expert annotations. Among the models analyzed, DenseNet121 shows the highest result, reaching an accuracy of 99.75% and having the values of precision, recall, and F1-score close to 0.998.
No takes yet. Share an insight, caveat, or question.
Mohapatra et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: