How to Reduce False Positives in Scanning

Today, Radiology published our latest study on breast cancer. This work, led by Felipe Oviedo Perhavec from Microsoft’s AI for Good Lab and Savannah Partridge (UW/Fred Hutch) in collaboration with researchers from Fred Hutch , University of Washington, University of Kaiserslautern-Landau, and the Technical University of Berlin, explores how AI can improve the accuracy and trustworthiness of breast cancer screening. We focused on a key challenge: MRI is an incredibly sensitive screening tool, especially for high-risk women—but it generates far too many false positives, leading to anxiety, unnecessary procedures, and higher costs. Our model, FCDD, takes a different approach. Rather than trying to learn what cancer looks like, it learns what normal looks like and flags what doesn’t. In a dataset of over 9,700 breast MRI exams—including real-world screening scenarios—our model: Doubled the positive predictive value vs. traditional models Reduced false positives by 25% Matched radiologists’ annotations with 92% accuracy Generalized well across multiple institutions without retraining What’s more, the model produces visual heatmaps that help radiologists see and understand why something was flagged—supporting trust, transparency, and adoption. We’ve made the code and methodology open to the research community. You can read the full paper in Radiology https://lnkd.in/gc82kXPN AI won't replace radiologists—but it can sharpen their tools, reduce false alarms, and help save lives.

Google DeepMind Introduces CoDoC: Revolutionizing Medical AI Diagnostics Open-source code available on GitHub for further research and development. Functionality: CoDoC decides when to utilize AI predictions or when human clinical judgment should take over for interpreting medical images. Achievements: Significantly lowers false positives in mammography by 25% while preserving true positives. Reduces clinician workload by potentially two-thirds in simulated environments. How it works: AI Prediction: An existing AI model, already designed to analyze medical images (like mammograms or chest X-rays), provides a confidence score for each image it assesses. This score ranges from 0 (no disease) to 1 (disease definitely present). Clinician's Interpretation: A clinician also reviews the same image and provides their diagnosis or interpretation. Ground Truth: Later, the actual diagnosis is confirmed through methods like biopsies or other clinical follow-ups. This is the definitive answer or ground truth. Training CoDoC:Learning Phase: CoDoC uses these three pieces of information for each case in a dataset to learn. It compares:The AI's confidence score with the clinician's interpretation. Both of these with the ground truth to understand when each is correct or incorrect. Establishing Accuracy: Through this comparison, CoDoC learns to recognize patterns where the AI or the clinician is more accurate based on the AI's confidence levels. Decision Making: Real-time Assessment: Once trained, when a new medical image comes in: The AI analyzes it and gives a confidence score. This score is fed into CoDoC. Decision Point: CoDoC then decides:,If the AI's confidence score falls in a range where it's known to be more accurate than the clinician, CoDoC might recommend trusting the AI's diagnosis. If the confidence score suggests the AI might be less accurate, or if the situation is ambiguous, CoDoC advises deferring to a human clinician. Outcome: Enhanced Workflow: This process aims to: Reduce unnecessary tests or false positives by leveraging AI when it's most reliable. Ensure human expertise is involved when necessary, enhancing diagnostic accuracy. Application: Deployment: CoDoC can be deployed in a clinical setting where both AI tools and human clinicians are available, guiding the workflow for better resource allocation and diagnostic outcomes. https://lnkd.in/g9nUh5Kq

**5 Key Lessons from Automating Security Decisions with Arcanna Ai & Google Siemplify** In the past year, I've integrated Arcanna AI with our security operations, significantly improving our response time and accuracy. Here are five insights that I gained from this experience, each saving time and reducing effort. If you're involved in security operations, these insights could potentially streamline your workflows. *Lesson 1: Embrace Integration for Efficiency* A major challenge in security operations is handling vast amounts of data. Many teams try to manage manually, which leads to delays. The reality is, integrating platforms like Arcana AI with systems such as Siemplify SOAR can transform your operations. By automating decision-making, we cut our incident response time by 30%. Ensure key integrations are up to date. This maximizes the systems' potential and effectiveness. *Lesson 2: Use AI for Decision Support* Security teams often rely solely on human judgment, which can be inconsistent. Arcanna AI provides consistent decision support based on accumulated data and learning. When we started, initial skepticism faded as the reliability became evident through reduced false positives. Implement AI-based decision support. It enhances accuracy and confidence in security measures. *Lesson 3: Provide Continuous Feedback for Improvement* A common misconception is that AI models are static and unchanging. In reality, providing feedback improves AI models significantly. Initially, our models struggled with identifying complex threats. With continuous feedback, detection rates improved. Keep offering feedback to the AI to refine its decision-making capabilities. *Lesson 4: Prioritize Retraining for Relevance* AI solutions can become outdated without regular updates, leading to ineffective responses. Regular training ensures the AI evolves with new data inputs. The changes we introduced increased the model's precision by almost 40%. Schedule regular retraining sessions. This maintains the relevance and efficiency of your AI tools. *Lesson 5: Prepare for Initial Learning Curves* New implementations can face resistance due to unfamiliarity. However, after initial adjustments, Arcana AI's integration became a crucial part of our team. The initial phase took time, but results soon aligned with expectations. Anticipate an initial learning period. The benefits solidify over time as familiarity grows. These lessons highlight the potential of integrating AI tools like Arcanna into security operations. Trust in technology, ongoing improvements, and adapting processes are key to maximizing performance. Ready to enhance your security operations? Start by integrating and trusting decision intelligence platforms within your workflow. https://lnkd.in/eNJFX59k

Uncover the power of Neuro-symbolic AI in Financial Fraud Detection. This week's deep dive explores how combining neural networks with symbolic reasoning is revolutionizing fraud prevention, achieving 96.5% accuracy and processing 100,000 transactions per second! 🎯 Featured insights: Architecture breakdowns, implementation strategies, and how this hybrid approach reduces false positives by 76%. Essential reading for fintech professionals, AI engineers, and security architects. #TechInsights #AI #FinTech #FraudDetection #MachineLearning #Finance #Innovation #Banking

If you're tired of patching 10,000 vulnerabilities that don't matter… This post will show you how to fix that. Because that’s exactly why I built Strobes. Back in 2019, I stood in front of a client’s dashboard with 90,000 open findings. They had best-in-class tools. They had a strong security team. They were still vulnerable in all the wrong places. Because when everything looks urgent, nothing gets fixed in time. That was the moment I realized: Enterprises don’t need another scanner. They need a system that helps them see clearly before they act quickly. So what does Strobes CTEM actually do? It helps security teams: A) Go from noisy dashboards to a single risk view B) Go from endless CVSS scores to risk**-prioritized threats** C) Go from ticket backlogs to automated remediation workflows We unify → ASM + RBVM and Pentesting into one continuous motion. And we work with some of the most security-conscious enterprises in the world to help them: 1. Prioritize the 3% of vulnerabilities that actually matter 2. Hit 98% SLA compliance across the board 3. Reduce remediation time by 60% 4. Drop false positives by 82% What does that look like in action? An enterprise e-commerce platform Came to us buried under 55,000 vulnerabilities. Four scanners. Multiple teams. Total chaos. We brought everything into Strobes CTEM and? ↳ 67% faster remediation → 82% fewer false positives → 98% SLA compliance That didn’t happen because they worked harder. It happened because they finally knew where to aim. This is what Strobes was built for. Not more dashboards. Not more alerts. Just the truth about what will break you… And the power to fix it before it does.