BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:IBIIS & AIMI Seminar - "GREEN: Generative Radiology Report Eva
 luation and Error Notation" & " Leveraging Patch-Level Representation Lear
 ning with Vision Transformer for Prostate Cancer Foundation Models"
X-WR-TIMEZONE:Pacific Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20261006T034325Z
UID:tag:localist.com\,2008:EventInstance_47489673081872
DTSTART:20240918T190000Z
DTEND:20240918T200000Z
DESCRIPTION:Sophie Ostmeier\, MD\nPostdoctoral Scholar\nDepartment of Radio
 logy\nStanford School of Medicine\n\nTitle: GREEN: Generative Radiology Re
 port Evaluation and Error Notation\n\nAbstract\nEvaluating radiology repor
 ts is a challenging problem as factual correctness is extremely important 
 due to the need for accurate medical communication about medical images. E
 xisting automatic evaluation metrics either suffer from failing to conside
 r factual correctness (e.g.\, BLEU and ROUGE) or are limited in their inte
 rpretability (e.g.\, F1CheXpert and F1RadGraph). In this paper\, we introd
 uce GREEN (Generative Radiology Report Evaluation and Error Notation)\, a 
 radiology report generation metric that leverages the natural language und
 erstanding of language models to identify and explain clinically significa
 nt errors in candidate reports\, both quantitatively and qualitatively. Co
 mpared to current metrics\, GREEN offers: 1) a score aligned with expert p
 references\, 2) human interpretable explanations of clinically significant
  errors\, enabling feedback loops with end-users\, and 3) a lightweight op
 en-source method that reaches the performance of commercial counterparts. 
 We validate our GREEN metric by comparing it to GPT-4\, as well as to erro
 r counts of 6 experts and preferences of 2 experts. Our method demonstrate
 s not only higher correlation with expert error counts\, but simultaneousl
 y higher alignment with expert preferences when compared to previous appro
 aches.\n\n\n\nJeong Hoon Lee\, PhD\nPostdoctoral Researcher\nDepartment of
  Radiology\nStanford School of Medicine\n\nTitle: Leveraging Patch-Level R
 epresentation Learning with Vision Transformer for Prostate Cancer Foundat
 ion Models\n\nAbstract: Recent advancements in self-supervised learning (S
 SL)\, emerging as an effective approach for imaging foundation models\, en
 able the effective pretraining of AI models across multiple domains withou
 t the need for labels. Despite the rapid advancements\, their application 
 in medical imaging remains challenging due to the subtle difference betwee
 n cancer and normal tissue. To address this limitation\, in this study\, w
 e propose an AI architecture ProViCNet that employs the vision transformer
  (ViT) based segmentation architecture with patch-level contrastive learni
 ng for better feature representation. We validated our model in prostate c
 ancer detection tasks using three types of magnetic resonance imaging (MRI
 ) across multiple centers. To evaluate the performance of feature represen
 tation in this model\, we performed downstream tasks with respect to Gleas
 on grade score and race prediction. Our model demonstrated significant per
 formance improvements compared to the state-of-the-art segmentation archit
 ectures. This study proposes a novel approach to developing foundation mod
 els for prostate cancer imaging overcoming SSL limitations.
LOCATION:Clark Center S360
SUMMARY:IBIIS & AIMI Seminar - "GREEN: Generative Radiology Report Evaluati
 on and Error Notation" & " Leveraging Patch-Level Representation Learning 
 with Vision Transformer for Prostate Cancer Foundation Models"
URL;VALUE=URI:https://events.stanford.edu/event/ibiis-aimi-seminar-green-ge
 nerative-radiology-report-evaluation-and-error-notation-a-foundation-ai-mo
 del-for-prostate-cancer-imaging-and-its-application-for-recurrence-predict
 ion-from-focal-ablation
CATEGORIES:Class/Seminar
END:VEVENT
END:VCALENDAR