Le Lu

RESEARCH A deep learning model shows that acute aortic syndrome can be diagnosed directly from noncontrast CT, increasing accuracy & decreasing time to #diagnosis. #artificial_intelligence nature.com/articles/s4159…

A compelling perspective trial conducted with clearly positive patient outcomes

[KOREA AI SUMMIT 2024] Cluster 2 : Dr. Le Lu (Alibaba DAMO Academy) youtu.be/EvSQNRUC6KA?si… via @YouTube

How far can we go with vision alone? Excited to reveal our Large Vision Model! Trained with 420B tokens, effective scalability, and enabling new avenues in vision tasks! (1/N) Kudos to @younggeng @Karttikeya_m @_amirbar, @YuilleAlan Trevor Darrell @JitendraMalikCV Alyosha Efros!

Oxford HPB Journal Club Superb, practice-changing study ushering the way for non-contrast CT for the detection of pancreatic cancer in asymptomatic individuals. Training of the PANDA AI tool which can detect PDaC with exceptional diagnostic accuracy. nature.com/articles/s4159…

If you ever wonder why immunotherapy works for some patients but not others, our new paper in npj Precision Oncology, "Biology-aware mutation-based deep learning for outcome prediction of cancer immunotherapy with ICIs", should be useful 4 u. rdcu.be/dqlRr

@JorgKleeff ppv/precision is 0.56 in two classes of cancer (all stages) versus non-cancer. see table 14 in supplementary ... We expect there will be more than 20% patients with resectable PDAC when there is an effective screening tool available, but only real number will tell ... later.

AI and imaging-based cancer screening: getting ready for prime time rdcu.be/drzof nature.com/articles/s4159…

some lights on AUC values. In screening we are essentially solving a highly unbalanced problem of cancer and non-cancer patients (AUC is easier to be higher) Pancreatic cancer prevalence is 13 out of 100K, for screening population maybe 30 out of 100K. PPV is 0.56 (table 14).

Meta-information-aware Dual-path Transformer for Differential Diagnosis of Multi-type Pancreatic Lesions in Multi-phase CT, IPMI 2023 link.springer.com/chapter/10.100…

Devil is in the Queries: Advancing Mask Transformers for Real-world Medical Image Segmentation and Out-of-Distribution Localization, CVPR 2023 (highlights) computer.org/csdl/proceedin…

Continual Segment: Towards a Single, Unified and Non-forgetting Continual Segmentation Model of 143 Whole-body Organs in CT Scans, ICCV 2023 openaccess.thecvf.com/content/ICCV20…

Deep Learning for Fully Automated Prediction of Overall Survival in Patients Undergoing Resection for Pancreatic Cancer: A Retrospective Multicenter Study, Annals of Surgery 278(1):p e68-e79, July 2023. journals.lww.com/annalsofsurger…

Comprehensive and clinically accurate head and neck cancer organs-at-risk delineation on a multi-institutional study, Nature Communications volume 13, Article number: 6137 (2022) nature.com/articles/s4146…

A flexible three-dimensional heterophase computed tomography hepatocellular carcinoma detection algorithm for generalizable and practical screening, Hepatol Commun. 2022 Oct; 6(10): 2901–2913. Published online 2022 Jul 19. doi: 10.1002/hep4.2029

Artificial Intelligence to Predict Lymph Node Metastasis on CT in Pancreatic Ductal Adenocarcinoma, Radiology, January 2023 pubs.rsna.org/doi/abs/10.114…

We have some other interesting work cooking in house ... Combating cancer using computing tools is a long game. CancerUniT: Towards a Single Unified Model for Effective Detection, Segmentation, and Diagnosis of Eight Major Cancers Using a Large Collection of CT Scans, ICCV 2023

PS. I get worried when I see an AUC ~0.99 and 99.9% specificity! This model went through many stages of testing and validation. Solid accompanying editorial nature.com/articles/s4159… by @JorgKleeff free access rdcu.be/drzof

A very impressive study for #AI picking up pancreatic cancer from non-contrast CT with 93% sensitivity, 99.9% specificity, AUC ~0.99, including from chest CT scans performed Covid. nature.com/articles/s4159… @NatureMedicine

usually i’m so good at tuning out current events, but when any founder who helped build a company from scratch is on the line for getting kicked out. kills me. just too great a loss. i feel for ilya. the only way openai was able to recruit him out of google brain was with the promise they would be different. they wouldn’t be daddied by any large tech company. that they’d give it all away for free. imagine being the guy whose key insights and research have made this technology possible? you thought you were giving that talent and those discoveries to the right people? so sad. i feel for them all. i really do.

with sam and greg ousted yesterday my initial thought yesterday was “shit they have discovered AGI” and the only person who was capable of understanding it was ilya sutskever. now it’s seems even more an more plausible that is the case. to understand the amount of stress and pressure he must have been under i urge you to watch this short documentary. this was recorded during the development of chatGPT. but the sincere stream of thought about the greater good AND the bad that will come from AI might put it more clear to why this might have happened. my thesis, they have had AGI internally for a while, and Ilya, andrej karpathy might have been the first ones to grasp what they have made. to be clear, geoff hinton regrets his life’s work (the work on AI) due to him worry about all the bad it can be used for. maybe ilya is under similar pressure. to much for one person to bear. regardless, we will know more soon. i feel for all involved parties, and for the community at large. this is not great.