2026-10-09T19:51:42Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33436735 2026-09-17T18:32:50Z category_26275 category_28840 category_28849 category_28879 category_28885 category_28888 portal_549 item_type_3 month_year_09_2026

PitVQA-Anticipation dataset with images and question-answer pairs Shreyas Dhake (22344298) Jiayuan Huang (5767256) Runlong He (19766379) Danyal Z. Khan (11029082) Evangelos Mazomenos (6761597) Sophia Bano (6773012) Hani Marcus (6766427) Danail Stoyanov (6778706) Matt Clarkson (6771092) Mobarak Hoque (13180839) Biomedical imaging Intelligent robotics Natural language processing Computer vision Image processing Multimodal analysis and synthesis visual question answering (VQA) large language models in medicine Large language models (LLMs) in healthcare Vision Language Models (VLMs) Pituitary surgery Artificial Intelligence (AI) · Robotic surgery · Intraoperative decision-making · Surgical automation · Minimally invasive surgery Surgical data science artificial intelligence analysis <p dir="ltr"><b>PitVQA-Anticipation</b> is a visual question-answering dataset for forward-looking reasoning in endoscopic pituitary surgery. It is derived from 25 surgical videos recorded at the National Hospital for Neurology and Neurosurgery in London, United Kingdom. The procedures were recorded using a high-definition Karl Storz endoscope at 720p resolution and comprise approximately 33.5 hours of operative video. Frames were extracted at 1 frame per second, with blurred or occluded frames removed, resulting in 109,173 usable frames.Each frame is associated with exactly seven question–answer pairs covering four anticipation tasks: forecasting future surgical phases, predicting the next operative step, identifying instruments required for the next step, and estimating the remaining duration of the current step, current phase and complete procedure. The full frame-level dataset contains 764,211 question–answer pairs. Its target annotations cover 39 categories, comprising 4 future phases, 14 future steps, 18 instruments and 3 remaining-time targets.The source videos were annotated by two neurosurgical residents with experience in pituitary surgery and reviewed by an attending neurosurgeon. According to the source dataset documentation, all patients provided informed consent and the study was registered with the local governance committee.<br><br><b>Acknowledgement</b></p><p dir="ltr"><br></p><ul><li>PitVQA images are derived from MICCAI PitVis challenge:</li><li>PitVis Paper: https://arxiv.org/abs/2409.01184</li><li>PitVis Challenge: https://www.synapse.org/Synapse:syn51232283</li><li>PitVis Dataset: https://doi.org/10.5522/04/26531686</li></ul><p><br></p><p><br></p><p dir="ltr">If you use this code for your research, please cite our paper.</p><p dir="ltr">https://doi.org/10.1007/s11548-026-03784-z</p> 2026-09-17T18:32:50Z Dataset Dataset 10.5522/04/33436735.v1 https://figshare.com/articles/dataset/PitVQA-Anticipation_dataset_with_images_and_question-answer_pairs/33436735 CC BY-NC-ND 4.0