Deep Learning in Medical Imaging: A Case Study on Lung Tissue Classification

The paper has several important inconsistencies that should be clarified. First, the authorship information is not consistent across EAI records: the published PDF lists seven authors, while another EAI record reportedly lists only five and omits Hritwik Ghosh and Irfan Sadiq Rahat. This is especially unusual because Hritwik Ghosh is identified as the corresponding author in the PDF.

There is also an apparent authorship error in Reference [16], where one of the authors seems to have been replaced by Abdus Sobur, who is also an author of the present paper. The literature review also includes several weakly related citations on topics such as water quality, Walmart data, stock prediction and cyberbullying, which do not clearly support a lung histopathology study.

The methodological inconsistencies are even more serious. The dataset description strongly resembles LC25000, but the dataset is not clearly identified or properly cited. The paper states that 9,000 images were split 70/15/15, meaning the test set should contain 1,350 images. However, some confusion matrices contain about 3,000 test samples, which directly conflicts with the stated split.

The paper also claims that EfficientNetB5 achieved 100% accuracy, but its own confusion matrix shows five misclassified images, giving about 99.83% accuracy rather than 100%. Figure 12 also appears mislabeled, and several performance metrics promised in the abstract, such as precision, recall and F1-score, are not fully reported.