With the popularity of artificial intelligence, specifically in deep learning domain, there are a large host of neural models due to various learning techniques like supervised, transfer learning amongst various others techniques allow to automate diagnosis; thereby, reducing workload on healthcare professionals or lack of workforce availability and reducing the chances of mis-identification or diagnosis.
 In order to ensure trust among both professionals working with the automated system and the patients that are being diagnosed are efficient, a need is there for employing interpretation of the model used, while performing its task.
The transformer model used to perform diagnosis for the gastrointestinal tract by multi-class classification of anomalies using wireless capsule endoscopy images from Kvasir dataset and a visualization technique as grad-cam which allows to determine the key regions (i.e eliminate the black box nature of neural network) within the GiT that the model uses to classify.