Pitvqa: Image-grounded Text Embedding LLM For Visual Question Answering In Pituitary Surgery · The Large Language Model Bible Contribute to LLM-Bible

The Large Language Model Bible

Research on Large Language Models. Maintained by Sean Moran.

Contact Sean Moran about this survey or website. Made with Jekyll and Hyde.

Pitvqa: Image-grounded Text Embedding LLM For Visual Question Answering In Pituitary Surgery

He Runlong, Xu Mengya, Das Adrito, Khan Danyal Z., Bano Sophia, Marcus Hani J., Stoyanov Danail, Clarkson Matthew J., Islam Mobarakol. Arxiv 2024

[Paper] [Code]    
Applications Attention Mechanism GPT Has Code Merging Model Architecture Multimodal Models RAG Reinforcement Learning

Visual Question Answering (VQA) within the surgical domain, utilizing Large Language Models (LLMs), offers a distinct opportunity to improve intra-operative decision-making and facilitate intuitive surgeon-AI interaction. However, the development of LLMs for surgical VQA is hindered by the scarcity of diverse and extensive datasets with complex reasoning tasks. Moreover, contextual fusion of the image and text modalities remains an open research challenge due to the inherent differences between these two types of information and the complexity involved in aligning them. This paper introduces PitVQA, a novel dataset specifically designed for VQA in endonasal pituitary surgery and PitVQA-Net, an adaptation of the GPT2 with a novel image-grounded text embedding for surgical VQA. PitVQA comprises 25 procedural videos and a rich collection of question-answer pairs spanning crucial surgical aspects such as phase and step recognition, context understanding, tool detection and localization, and tool-tissue interactions. PitVQA-Net consists of a novel image-grounded text embedding that projects image and text features into a shared embedding space and GPT2 Backbone with an excitation block classification head to generate contextually relevant answers within the complex domain of endonasal pituitary surgery. Our image-grounded text embedding leverages joint embedding, cross-attention and contextual representation to understand the contextual relationship between questions and surgical images. We demonstrate the effectiveness of PitVQA-Net on both the PitVQA and the publicly available EndoVis18-VQA dataset, achieving improvements in balanced accuracy of 8% and 9% over the most recent baselines, respectively. Our code and dataset is available at https://github.com/mobarakol/PitVQA.

Similar Work