John Snow Labs’ Post

Don’t miss our upcoming webinar ‘De-identification of Medical Images in DICOM Format’ on November 13th at 2pm ET by Jose Pablo Alberto Andreotti, our Data Scientist. De-identifying medical records is key to unlocking valuable data for privacy, compliance, research, and reducing data breaches. DICOM, a standard for exchanging medical images (e.g., CT, MRI, ultrasound), poses anonymization challenges due to: ♢ Sensitive info “burned” into images, requiring computer vision/OCR ♢ Unstructured text in metadata ♢ Thousands of DICOM file variants and metadata fields ♢ Files containing thousands of images in various resolutions ♢ Modality-specific nuances (MRI vs. CT vs. ultrasound) This session presents a scalable, high-accuracy solution for processing and de-identifying DICOM files using John Snow Labs' Visual NLP, with live demos and infrastructure insights for scaling pipelines. Read more: https://lnkd.in/dd--d8NN #LargeLanguageModels #MedicalAIApplications #AIinHealthcare #LLMs #HealthcareLLMs #GenerativeAI #HealthcareAI #MedicalLLMs #nlp

Join ‘De-identification of Medical Images in DICOM Format’ webinar on November 13th @ 2pm ET by Jose Pablo Alberto Andreotti. De-identifying medical records is key to unlocking valuable data for privacy, compliance, research, and reducing data breaches. DICOM, a standard for exchanging medical images (e.g., CT, MRI, ultrasound), poses anonymization challenges due to: - Sensitive info “burned” into images, requiring computer vision/OCR - Unstructured text in metadata - Thousands of DICOM file variants and metadata fields - Files containing thousands of images in various resolutions - Modality-specific nuances (MRI vs. CT vs. ultrasound) This session presents a scalable, high-accuracy solution for processing and de-identifying DICOM files using John Snow Labs' Visual NLP, with live demos and infrastructure insights for scaling pipelines. Register now: https://lnkd.in/dd--d8NN #LargeLanguageModels #MedicalAIApplications #AIinHealthcare #LLMs #HealthcareLLMs #GenerativeAI #HealthcareAI #MedicalLLMs #nlp

Join ‘De-identification of Medical Images in DICOM Format’ webinar on November 13th @ 2pm ET by Alberto Andreotti. De-identifying medical records is key to unlocking valuable data for privacy, compliance, research, and reducing data breaches. DICOM, a standard for exchanging medical images (e.g., CT, MRI, ultrasound), poses anonymization challenges due to: - Sensitive info “burned” into images, requiring computer vision/OCR - Unstructured text in metadata - Thousands of DICOM file variants and metadata fields - Files containing thousands of images in various resolutions - Modality-specific nuances (MRI vs. CT vs. ultrasound) This session presents a scalable, high-accuracy solution for processing and de-identifying DICOM files using John Snow Labs' Visual NLP, with live demos and infrastructure insights for scaling pipelines. Register now: https://hubs.li/Q02VyPfR0 #LargeLanguageModels #MedicalAIApplications #AIinHealthcare #LLMs #HealthcareLLMs #GenerativeAI #HealthcareAI #MedicalLLMs #nlp

Join Our Upcoming Webinar: De-Identification of Medical Images in DICOM Format. Date: November 13, 2024 Time: 2:00 PM ET The de-identification of medical records is essential for privacy, compliance, and enabling medical research. When it comes to medical images, particularly in DICOM format, the challenges multiply. From sensitive information burned into images to metadata containing unstructured text, accurately anonymizing DICOM files requires advanced solutions. In this live webinar, we’ll address these challenges and showcase how John Snow Labs' Visual NLP is tackling the complexities of de-identifying medical images, including: - De-identifying images and metadata across various clinical modalities (MRI, CT, Ultrasound, etc.) - Using computer vision and OCR for image-based anonymization - Scaling pipelines to process large DICOM datasets efficiently What to Expect: - Live demos and code for DICOM processing - A deep dive into the scalability and accuracy of enterprise-grade de-identification solutions - Insights into how to handle heavy workloads and multiple image formats- Whether you’re involved in medical research or working with healthcare data, this session will provide valuable insights into how you can safely and efficiently process medical images while maintaining compliance. Register Now: https://lnkd.in/dG7x-UH8 #DeIdentification #MedicalImages #DICOM #HealthcareAI #NLP #ComputerVision #MedicalData #AI #HealthTech #DataPrivacy #JohnSnowLabs

I am pleased to announce that two of our papers have been accepted at the 2nd International Workshop on Multimodal Content Analysis for Social Good (MM4SG), co-located with the IEEE International Conference on Data Mining (ICDM 2024), a CORE Rank A* conference. 1️⃣ Optimized Biomedical Question-Answering Services with LLM and Multi-BERT Integration This paper introduces a novel architecture that integrates Large Language Models (LLMs) and multiple BERT configurations to improve the accuracy and efficiency of biomedical question-answering systems. The approach is designed to support healthcare professionals by providing reliable data-driven insights for better decision-making. 2️⃣ Fine-Tuning LLMs for Reliable Medical Question-Answering Services This work focuses on enhancing the performance of medical question-answering services by fine-tuning advanced LLMs like LLaMA-2 and Mistral. Using techniques such as rsDoRA+ and ReRAG, we aim to deliver fast and accurate medical responses, ultimately contributing to improved healthcare outcomes. These contributions reflect our ongoing commitment to leveraging AI for social good and advancing the field of healthcare technology. #ICDM2024 #COREAStar #BiomedicalQA #Healthcare #AI #MachineLearning #NLP #Research

Labellerr helps us to scale our image annotation operation - Eric, Senior Scientist at Foss Learn from our customers how Labellerr helps them solve the challenges around preparing training datasets for vision/NLP/LLM model training. Here, Eric, Senior scientist at Foss Analytics is sharing his experience working on our tool and how it helps him to solve the challenges in his image instance segmentation annotation project. Here he shared how Labellerr helped his team scale annotation throughput which was not possible before. #computervision #instances #segmentation #dataannotation #imageanalysis #testimonial #aisaas #artificialinteligence #ArtificialIntelligence #AI #MachineLearning #DeepLearning #DataScience #AIResearch #AICommunity #AIDevelopment #NeuralNetworks #AItechnology #BigData #TechInnovation #AIethics #FutureTech #Automation #Robotics #NLP  #ComputerVision #SmartTech #labellerr

A complex field like healthcare provides many opportunities for machine learning and generative AI applications. In practical terms, this often involves the art of building bridges between the possible and the feasible – while technical tools exist that make fancy applications possible, only a subset of these are feasible at a non-profit healthcare center within regulatory and resource constraints. In this, Sonali discusses healthcare applications using a multi-faceted approach, including domain context and practical limitations along with the technical set-up. Adaptability is key, with tools and problems judiciously chosen for meaningful and successful applications. Simplest natural language processing methods can still add value, projects that do not involve protected health information can serve as a sandbox while meeting important business needs, and sophisticated generative artificial intelligence techniques can be combined with traditional machine learning methods for maximal impact. This session was presented by Sonali Tamhankar – Sr. Clinical Data Scientist, Fred Hutchinson Cancer Center at #Healthcare #NLPSummit 2024 Watch an entire video here: https://lnkd.in/dVyqMSMa #LargeLanguageModels #MedicalAIApplications, #AIinHealthcare #LLMs #HealthcareLLMs #GenerativeAI #HealthcareAI #MedicalLLMs #nlp

💡 A review of "The Evolution of Multimodal Model Architectures" This paper identifies and characterizes four prevalent multimodal model architectural patterns, providing a systematic categorization that aids in monitoring developments in the multimodal domain. It distinguishes four specific architecture types based on their methods of integrating multimodal inputs: ✅ Type A: Uses standard cross-attention for deep fusion within the model's internal layers. ✅ Type B: Utilizes custom-designed layers for deep fusion within the model's internal layers. ✅ Type C: Employs modality-specific encoders for early fusion at the input stage. ✅ Type D: Leverages tokenizers to process modalities at the input stage. 🔗 Paper: https://lnkd.in/e-za-MCS #Ai #MachineLearning #DeepLearning #ComputerVision #NLP

Persivia Specialty Care Model gives you the freedom to deliver the best care for your patients with complex needs your way. We use machine learning, NLP, Predictive AI and advanced technology to empower you to make the best decisions for your patients. By harnessing extensive patient data from diverse sources and leveraging a robust data fabric enriched with thousands of data elements, AI and deep clinical knowledge, Persivia Specialty Care Model covers all programs across all available datasets to arrive at insights in real-time, personalized at the point of care. Join the Conversation at America's Physician Groups Annual Fall conference 2024, happening today! #APG #specialtycare #persivia #carespace #NLP #AI

Really exciting news: we got 4 papers accepted at #EMNLP2024 that belong to the "Multimodality and Language Grounding to Vision, Robotics and Beyond" track. Looking forward to meeting everyone who is attending EMNLP in Miami this year. Feel free to ping me to discuss Multimodal Learning and Embodied AI! List of accepted papers below: 1) "AlanaVLM: A Multimodal Embodied AI Foundation Model for Egocentric Video Understanding" (ArXiv: https://lnkd.in/dMrRUZUb) Authors: Alessandro Suglia Claudio Greco Katie Baker Jose L. Part Ioannis Papaioannou Arash Eshghi Ioannis Konstas Oliver Lemon TL;DR: We present a fine-tuning recipe to build VLMs able to understand egocentric videos and generate answers to visual queries. 2) "Shaking Up VLMs: Comparing Transformers and Structured State Space Models for Vision & Language Modeling" (ArXiv: https://lnkd.in/eiJpMgNM) Authors: George Pantazopoulos Malvina Nikandrou Alessandro Suglia Oliver Lemon Arash Eshghi TL;DR: Detailed comparison between Transformers-based and SSM-based Vision and Language models which allowed us to uncover some serious problems in current Multimodal #GenerativeAI models. 3) "Repairs in a Block World: A New Benchmark for Handling User Corrections with Multi-Modal Language Models" (on ArXiv soon) Authors: Javier Chiyah-Garcia Arash Eshghi Alessandro Suglia TL;DR: We explore VLMs' capabilities (including OpenAI GPT-4o) to process third-position repairs in a situated instruction-following task with a new dataset, and compare them with human participants with an in-person study. 4) "Investigating the Role of Instruction Variety and Task Difficulty in Robotic Manipulation Tasks" (ArXiv: https://lnkd.in/ebwv8439) Authors: Amit Parekh Nikolas Vitsakis, Alessandro Suglia, Ioannis Konstas TL;DR: Multimodal models are resilient to nonsensical instructions but they struggle with more common observational changes indicating a need for improvements in evaluating generalisation capabilities both inside and outside the domain. Heriot-Watt University The National Robotarium #NLProc #GenerativeAI #AI #ML #Robotics