LightGastroFormer: a lightweight multi-resolution transformer for gastrointestinal disease classification

Prateek Singh, Sudhakar Singh, Manoj Kumar Shukla · 2026

Read LightGastroFormer: a lightweight multi-resolution transformer for gastrointestinal disease classification by Prateek Singh, Sudhakar Singh, Manoj Kumar Shukla (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13427751.txt.

Description

Singh, Singh, and Shukla built LightGastroFormer, a lightweight transformer network. It classifies gastrointestinal diseases from endoscopy and capsule images. It stays small enough for hospital use in practice. Existing deep learning systems often miss either fine local lesion detail or broad context. They also struggle on the field's large, heavily imbalanced image sets.

The model combines a multi-resolution patch tokenizer, lean self-attention, and a gated feed-forward network. That mix catches small lesions without heavy math. It scored 0.94 accuracy on the Kvasir v1 benchmark and 0.95 on Kvasir v2 in the reported tests. On the highly imbalanced Kvasir-Capsule set it reached 0.97 accuracy, with a matching 0.97 F1-score. It did so with no special class-balancing tricks. The whole network carries only 6.42 million trainable parameters. Ablation studies showed each component helps.

Catalog record

CREATORPrateek Singh, Sudhakar Singh, Manoj Kumar Shukla
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE82 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC13427751

Browse the full library · About the Tower of Babel