Suggestions
Dominik Stosik, MD, PhD
| AI in HealthTech from Stanford | CEO & Founder at attained.ai | PhD in Pharmacology | Medical Doctor | US Patent Holder |
Dominik Stosik is a medical professional with a background in pharmacology, holding both an MD and a PhD. He serves as the CEO of attained.ai and is also involved in radiology. In addition to his professional roles, he is a Young Club Committee Member at the European Society of Medical Imaging Informatics (EuSoMII) .1
Stosik actively engages in discussions on topics related to medical imaging and artificial intelligence, as evidenced by his contributions on LinkedIn, where he shares insights about teaching large language models and the importance of data in healthcare .123 His work reflects a commitment to advancing the field of medical imaging informatics through innovative approaches and collaboration.
Highlights
Introducing StylEx: Revolutionizing Medical Imaging with Generative AI - Google Research
StylEx, a groundbreaking generative AI model, is transforming the landscape of medical imaging by providing unprecedented insights into complex visual features. This innovative model is set to redefine how we understand and interpret medical images across various modalities.
Key Features:
-
Unified Framework: StylEx leverages the powerful StyleGAN2 architecture to generate counterfactual visualizations, elucidating attributes that influence medical imaging classifiers.
-
Interdisciplinary Expert Review: Attributes generated by StylEx are reviewed by a multidisciplinary panel, including clinicians, socio-technical experts, and social scientists, ensuring comprehensive evaluation and interpretation.
-
Broad Applicability: Applied to diverse imaging modalities like retinal fundus photographs, external eye photographs, and chest radiographs, StylEx has demonstrated its capability to identify both known and novel visual features.
-
Enhanced Explainability: By visualizing how changes in specific attributes affect classification outcomes, StylEx offers deep insights into AI decision-making processes, aiding in the identification of potential biases and confounders.
-
Scalable and Robust: The model’s ability to generate high-quality visual explanations across large datasets ensures its applicability in various clinical and research settings.
Applications and Impact:
StylEx is poised to revolutionize medical image analysis, providing a robust tool for researchers and clinicians. Its unique approach enables the discovery of new medical insights and improves our understanding of AI-based models. The integration of socio-ecological and structural determinants of health frameworks ensures that findings are not only scientifically rigorous but also socially and ethically grounded.
🖋️ Oran Lang, Doron Yaya-Stupp, Ilana Traynis, Heather Cole-Lewis, PhD, Chloe R. Bennett, Courtney R. Lyles, Charles Lau, Michael Irani, Christopher Semturs, Dale R. Webster, Greg S. Corrado, Avinatan Hassidim, Yossi Matias, Yun Liu, Naama Hammel, M.D, Boris Babenko
📄 Read the full paper here: https://lnkd.in/dx33UvvE
#AI #MachineLearning #MedicalAI #ImageAnalysis #HealthcareInnovation #DeepLearning #GenerativeAI #StylEx #MedicalImaging #Explainability #TechRevolution #FutureOfHealthcare #DataScience #HealthTech #ClinicalAI #MedicalResearch
Microsoft Research introduces BiomedParse, a biomedical foundation model for image parsing that can simultaneously perform segmentation, detection, and recognition for 82 object types across 9 imaging modalities.
BiomedParse is a new foundation model for holistic image analysis, integrating segmentation, detection, and recognition tasks. Traditionally handled separately, these tasks are now enhanced through BiomedParse’s interdependencies.
Key Features:
-
Unified Framework: BiomedParse conducts segmentation, detection, and recognition across 82 object types and 9 imaging modalities, improving accuracy and efficiency.
-
Semantic Learning with GPT-4: Harmonizes natural language descriptions with biomedical ontologies, creating a comprehensive dataset of 6.8 million image-mask-description triples.
-
State-of-the-Art Performance: Outperforms methods like MedSAM, achieving a Dice score of 0.857, 39.6% higher than the best-competing method.
-
Scalability with Text Prompts: Segments objects using text prompts alone, eliminating the need for bounding box annotations, making the model highly scalable.
-
Invalid Input Detection: Uses Kolmogorov-Smirnov test to detect and reject invalid inputs, enhancing reliability.
Applications and Impact:
BiomedParse will revolutionize biomedical image analysis, providing a robust, scalable tool for researchers and clinicians. Its ability to handle complex, irregular-shaped objects across various modalities makes it invaluable in fields like pathology and radiology. Continued advancements will lead to more efficient and accurate image-based biomedical discoveries.
✍️ Theodore Zhao, Yu Gu, Jianwei Yang, Naoto Usuyama, Ho Hin Lee, Tristan Naumann, Jianfeng Gao, Angela Crabtree, Brian Piening, Carlo Bifulco, Mu Wei, Hoifung Poon, Sheng Wang
📄 Read the full paper here: https://lnkd.in/dEP_cE2t
#AI #MachineLearning #BiomedicalAI #ImageParsing #Radiology #HealthcareInnovation #MedicalImaging #DeepLearning #GPT4 #TechRevolution #MedicalAI #Pathology #RadiologyTech #FutureOfHealthcare #DataScience
