Usual interstitial pneumonia (UIP) and nonspecific interstitial pneumonia (NSIP) shared similar patterns in medical images but had different treatments and outcomes. This study presented an automatically distinguishable tool, named ConvT, using deep learning methods to help the clinicians differentiate NSIP from UIP based on MRI images. ConvT employed a CNN feature extractor and a cross-view transformer to classify the biomedical patterns that indicate NSIP or UIP. The experiment in a dataset with over 417 MRI T2WI images showed ConvT had 97.62% accuracy and 0.9767 F1 scores in the classification task. The visualization of ConvT illustrated that the proposed model can correctly identify the specific patterns of both NSIP and UIP.