![[2602.16422] Automated Histopathology Report Generation via Pyramidal Feature Extraction and the UNI Foundation Model](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16422] Automated Histopathology Report Generation via Pyramidal Feature Extraction and the UNI Foundation Model
Summary
This article presents a novel framework for generating histopathology reports using a combination of a foundation model and a Transformer decoder, addressing challenges in processing gigapixel images.
Why It Matters
Automating histopathology report generation can significantly enhance diagnostic efficiency and accuracy in medical settings, reducing the burden on pathologists and improving patient outcomes. This research contributes to advancements in AI applications within healthcare.
Key Takeaways
- The proposed framework utilizes a hierarchical vision language model for report generation.
- Multi-resolution pyramidal patch selection is employed to handle large image data effectively.
- The use of BioGPT for tokenization enhances the representation of biomedical terminology.
- A retrieval-based verification step ensures the reliability of generated reports.
- This approach could streamline workflows in histopathology and improve diagnostic accuracy.
Electrical Engineering and Systems Science > Image and Video Processing arXiv:2602.16422 (eess) [Submitted on 18 Feb 2026] Title:Automated Histopathology Report Generation via Pyramidal Feature Extraction and the UNI Foundation Model Authors:Ahmet Halici, Ece Tugba Cebeci, Musa Balci, Mustafa Cini, Serkan Sokmen View a PDF of the paper titled Automated Histopathology Report Generation via Pyramidal Feature Extraction and the UNI Foundation Model, by Ahmet Halici and 4 other authors View PDF HTML (experimental) Abstract:Generating diagnostic text from histopathology whole slide images (WSIs) is challenging due to the gigapixel scale of the input and the requirement for precise, domain specific language. We propose a hierarchical vision language framework that combines a frozen pathology foundation model with a Transformer decoder for report generation. To make WSI processing tractable, we perform multi resolution pyramidal patch selection (downsampling factors 2^3 to 2^6) and remove background and artifacts using Laplacian variance and HSV based criteria. Patch features are extracted with the UNI Vision Transformer and projected to a 6 layer Transformer decoder that generates diagnostic text via cross attention. To better represent biomedical terminology, we tokenize the output using BioGPT. Finally, we add a retrieval based verification step that compares generated reports with a reference corpus using Sentence BERT embeddings; if a high similarity match is found, the gene...
