SandboxAQ's AI speeds up biomarker discovery with Axia Medicine and Stanford

I’m excited to share that our paper “Evaluation of Monte Carlo Dropout for Uncertainty Quantification in Multi-task Deep-Learning-Based Glioma Subtyping” has been accepted as a poster presentation at the UNSURE workshop at #MICCAI2025 happening next week in Daejeon, South Korea! 🇰🇷🎉 In this study, we systematically evaluated Monte Carlo Dropout — one of the most widely used approaches to approximate Bayesian inference in Deep Learning — as an uncertainty quantification method for glioma diagnosis. This task includes tumor segmentation as well as the prediction of IDH mutation status, 1p19q co-deletion status, and tumor grade. We also disentangled aleatoric and epistemic uncertainty to gain a deeper understanding of model confidence. 🔍 Key findings: • When appropriately tuned, Monte Carlo Dropout enhances model trustworthiness by providing a good balance between performance and calibration. • Uncertainty disentanglement offers valuable insights into model behavior, aligning with current challenges discussed in the neuro-oncology field, and providing initial evidence that AI has the potential to move closer to human-level understanding of our medical task. 🤖🧠 I would like to thank my co-authors Sebastian van der Voort, Carolin Pirkl, Sandeep Kaushik, Marion Smits and Stefan Klein for their valuable contributions to this work. 📄 Preprint: https://lnkd.in/e-qZTFjk If you’re attending #MICCAI2025, I’d love to connect! Let’s exchange ideas and insights. 🤝 See you in Daejeon next week! 🇰🇷

🎉 Excited to announce that our paper, “𝑬𝒏𝒉𝒂𝒏𝒄𝒊𝒏𝒈 𝑳𝒖𝒏𝒈 𝑪𝒂𝒏𝒄𝒆𝒓 𝑹𝒆𝒄𝒐𝒈𝒏𝒊𝒕𝒊𝒐𝒏 𝒕𝒉𝒓𝒐𝒖𝒈𝒉 𝑫𝒆𝒆𝒑 𝑳𝒆𝒂𝒓𝒏𝒊𝒏𝒈: 𝑨 𝑪𝒐𝒎𝒑𝒂𝒓𝒂𝒕𝒊𝒗𝒆 𝑨𝒏𝒂𝒍𝒚𝒔𝒊𝒔 𝒐𝒇 𝑪𝑻 𝒂𝒏𝒅 𝑯𝒊𝒔𝒕𝒐𝒑𝒂𝒕𝒉𝒐𝒍𝒐𝒈𝒊𝒄𝒂𝒍 𝑰𝒎𝒂𝒈𝒊𝒏𝒈”, has been accepted at the 5th International Conference on Applied Intelligence and Informatics (AII 2025) in 𝐖𝐚𝐬𝐡𝐢𝐧𝐠𝐭𝐨𝐧 𝐃.𝐂., 𝐔𝐒𝐀 🇺🇸! This work highlights how AI-driven medical imaging can support early detection of lung cancer. By integrating CT scans and histopathological images, we demonstrate how deep learning can improve diagnostic accuracy and strengthen clinical decision-making. 📖 Our paper will also be published in the 𝑺𝒑𝒓𝒊𝒏𝒈𝒆𝒓-𝑵𝒂𝒕𝒖𝒓𝒆 𝘊𝘊𝘐𝘚 𝘱𝘳𝘰𝘤𝘦𝘦𝘥𝘪𝘯𝘨𝘴 𝘴𝘦𝘳𝘪𝘦𝘴, ensuring global academic visibility. 🔹 Authors: Md. Ashraful Babu, 𝐍𝐚𝐝𝐢𝐦 𝐀𝐡𝐦𝐞𝐝, Zarif Wasif Bhuiyan, and Mufti Mahmud Looking forward to sharing our findings and engaging with the vibrant research community at this prestigious event! 🌍 #AII2025 #DeepLearning #MedicalImaging #LungCancerDetection #AI #HealthcareAI #Springer

What if you could get a glimpse of the future of your health, today? Researchers at EMBL-EBI, DKFZ German Cancer Research Center, and the University of Copenhagen (Københavns Universitet) have developed a generative AI model that uses large-scale health records to estimate how human health may change over time. Like a weather forecast, the model doesn’t say exactly what will happen, but it can estimate risks over time. For example, it could predict the chance of developing heart disease within the next year, much like forecasting the chance of rain tomorrow. Our model is not ready for clinical use, but it’s a proof of concept that AI can help us understand long-term health patterns. Such models are an important step towards more personalised and preventive care, and they could help healthcare systems better anticipate and plan for our future.  Watch to find out more. Ewan Birney Tom Fitzgerald EMBL DKFZ German Cancer Research Center University of Copenhagen (Københavns Universitet) #bioinformatics #AI #precisionmedicine #healthinnovation

We're published our paper on AI in Healthcare - an innovative piece of AI that can make accurate risk estimates for future disease using individual's medical history. Check out the paper here: https://lnkd.in/dM_Jk277 And the video short here.

If - like me - you sometimes find talk about AI models to be a mix of hot air and magical thinking, this study might be for you. I'd argue that Delphi-2M actually 'gets' population health. It performs well when tested on Danish data, despite being tested on highly selected UK data, it reproduces comobidity patterns and it can generate disease trajectories out of the box (the G in GPT) and much more. The important take-away for me is that even Delphi-2M cannot predict, say, when I will get prostate cancer, it might still be a useful model. As part of my research at Rockwool Fonden, we examine how models like Delphi-2M can help address important problems without necessitating extremely precise predictions at the level of the individual. And there are a lot of those types of problems worth solving. Very happy to be part of this work along with colleagues from (among others) EMBL, DKFZ German Cancer Research Center, Danmarks Statistik and Københavns Universitet - University of Copenhagen. I am also thankful for the patience of the Novo Nordisk Foundation. That approach to funding is IMHO essential if researchers are to take on these types of challenges.

Fascinating look into the future of AI and healthcare from European Bioinformatics Institute | EMBL-EBI. A proof of concept on how AI can help tackle health issues through predictive modelling. Great to see it being picked up widely in the BBC News and other main media sites. https://lnkd.in/e8gYQ4Jj

Very exciting news 👏 Researchers at European Bioinformatics Institute | EMBL-EBI, DKFZ German Cancer Research Center (including one of our alumni, Moritz Gerstung) and the University of Copenhagen (Københavns Universitet) have developed a generative AI model that could help understand long-term health patterns. We had the pleasure of interviewing Moritz recently for our EMBL Alumni Chats series and will be releasing this on our YouTube channel very soon 🎦 https://lnkd.in/eMKrDrDc

Scientists have developed a new AI tool called Delphi-2M that can predict a person's risk of over 1,000 diseases. The tool uses data from diagnoses, medical events, and lifestyle factors to forecast health changes for the next decade and beyond. It was created by experts from the EMBL, the DKFZ German Cancer Research Center, and the Københavns Universitet - University of Copenhagen The AI model was trained on anonymized patient data from the UK Biobank study and the Danish national patient registry. Read more: https://lnkd.in/gUsSKGip 📰 Subscribe to the Life AI Weekly Newsletter: https://lnkd.in/eC5-u69w #ai #artificialintelligence #ainews #biotech #healthcareai

A groundbreaking AI model, Delphi-2M, has been developed to forecast a person's risk of diseases across their life. This innovative model, created by teams at the European Molecular Biology Laboratory and the German Cancer Research Centre, has the potential to revolutionize the field of medicine by allowing doctors to predict if their patients are likely to get one of over 1,000 different conditions, including Alzheimer's disease, cancer, and heart attacks. The model was trained on data from 400,000 people from the UK Biobank and was validated on data from the remaining 100,000 people in the Biobank, as well as on Danish health records. The results show that Delphi-2M performed well in predicting diagnoses within five years of a previous one, with an average AUC value of 0.76 on British data and 0.67 on Danish data. The potential applications of Delphi-2M are vast, and it could help health authorities allocate budgets for disease areas that may need extra funds in the future. While real-world applications remain far off, the model's accuracy and potential to improve patient outcomes make it an exciting development in the field of AI and medicine. Key statistics: 📊 Delphi-2M was trained on data from 400,000 people from the UK Biobank 📊 The model was validated on data from the remaining 100,000 people in the Biobank and on Danish health records 📊 Delphi-2M performed well in predicting diagnoses within five years of a previous one, with an average AUC value of 0.76 on British data and 0.67 on Danish data 📊 The model has the potential to predict over 1,000 different conditions, including Alzheimer's disease, cancer, and heart attacks #AI #ArtificialIntelligence #MachineLearning #Healthcare #Medicine #Innovation #Technology 🔄 Share 👍 React 🌐 Visit www.aravind-r.com #AravindRaghunathan

Scientists have developed a new artificial intelligence tool that can predict your personal risk of more than 1,000 diseases, and forecast changes in health a decade in advance. The generative AI tool was custom-built by experts from the European Molecular Biology Laboratory (EMBL), the German Cancer Research Centre and the University of Copenhagen, using algorithmic concepts similar to those used in large language models (LLMs). It is one of the most comprehensive demonstrations to date of how generative AI can model human disease progression at scale, and was trained on data from two entirely separate healthcare systems. Details of the breakthrough were published in the journal Nature. Source in comments.