https://mdsynapse.org/ Medical AI, imaging informatics, and clinical workflow automation, by Dr. Bradley J. Erickson. en-us https://mdsynapse.org/apple-touch-icon.pnghttps://mdsynapse.org/ Thu, 05 Nov 2026 12:00:00 +0000 https://mdsynapse.org/blogposts/any-language-model-can-be-a-decision-model https://mdsynapse.org/blogposts/any-language-model-can-be-a-decision-model Thu, 05 Nov 2026 12:00:00 +0000 Bradley J. Erickson AI What a radiology benchmark taught me about small System 1 models and about getting decisions out of ordinary language models https://mdsynapse.org/blogposts/can-system-1-models-reduce-pdoom https://mdsynapse.org/blogposts/can-system-1-models-reduce-pdoom Wed, 07 Oct 2026 12:00:00 +0000 Bradley J. Erickson AI My two most recent posts addressed separate questions. The first examined whether artificial intelligence (AI), or in the term now more commonly used, superinte https://mdsynapse.org/blogposts/fast-thinking-for-ai https://mdsynapse.org/blogposts/fast-thinking-for-ai Mon, 21 Sep 2026 12:00:00 +0000 Bradley J. Erickson AI A dual-process account of AI architecture, and a first evaluation of Jev on note-level PHI detection across 1000 synthetic clinical notes. https://mdsynapse.org/blogposts/pdoom-and-pboom-is-ai-too-risky-part-1 https://mdsynapse.org/blogposts/pdoom-and-pboom-is-ai-too-risky-part-1 Sat, 12 Sep 2026 12:00:00 +0000 Bradley J. Erickson AI On September 9, 2026, Jacob Coxon, a pretraining researcher who had worked at both OpenAI and Anthropic, resigned and posted a thread on X stating that the two https://mdsynapse.org/blogposts/sample-efficiency-test-time-training https://mdsynapse.org/blogposts/sample-efficiency-test-time-training Tue, 01 Sep 2026 12:00:00 +0000 Bradley J. Erickson AI A small transformer trained from scratch at test time scored 44% on ARC-AGI-1 for 67 cents. What its ablations suggest about sample efficiency, priors, and… https://mdsynapse.org/blogposts/ai-is-like-medicine https://mdsynapse.org/blogposts/ai-is-like-medicine Sat, 29 Aug 2026 12:00:00 +0000 Bradley J. Erickson AI The word "AI" is now applied to nearly all parts of life and science, including nucleic-acid classifiers, chatbots, robotic surgical arms. The challenge is that https://mdsynapse.org/blogposts/ai-efficiency-radiology https://mdsynapse.org/blogposts/ai-efficiency-radiology Wed, 22 Jul 2026 12:00:00 +0000 Bradley J. Erickson Clinical Use A review of the published evidence for AI-associated efficiency gains in radiology, covering report drafting, worklist triage, and workforce claims, and the… https://mdsynapse.org/blogposts/automation-and-ai-optimal-synergy https://mdsynapse.org/blogposts/automation-and-ai-optimal-synergy Mon, 13 Jul 2026 12:00:00 +0000 Bradley J. Erickson Automation LLMs are probabilistic components with a distinct failure signature. We argue that a deterministic automation backbone and an explicit exception-handling loop… https://mdsynapse.org/reviews/multimodal-learning-healthcare-benchmark https://mdsynapse.org/reviews/multimodal-learning-healthcare-benchmark Sun, 26 Apr 2026 12:00:00 +0000 Bradley J. Erickson Multimodal AI A review of CareBench, a benchmark of 14 EHR and chest radiograph fusion methods: when fusion helps, why modality balancing matters, and the effect on fairness. https://mdsynapse.org/blogposts/bio-ai https://mdsynapse.org/blogposts/bio-ai Sat, 28 Feb 2026 12:00:00 +0000 Bradley J. Erickson AI A review of machine learning applied to protein design, gene editing, and molecular data storage, and the implications for medicine and computation. https://mdsynapse.org/blogposts/hybrid-ai-workflows https://mdsynapse.org/blogposts/hybrid-ai-workflows Sat, 21 Feb 2026 12:00:00 +0000 Bradley J. Erickson Clinical Use Error compounds multiplicatively across sequential agent steps, which constrains the use of purely agentic workflows in clinical care. https://mdsynapse.org/blogposts/uncertainty-quantification https://mdsynapse.org/blogposts/uncertainty-quantification Thu, 01 Jan 2026 12:00:00 +0000 Bradley J. Erickson AI Most FDA-cleared AI triage tools output a binary alert without any confidence estimate. We review what uncertainty quantification measures, why calibration… https://mdsynapse.org/blogposts/foundation-models https://mdsynapse.org/blogposts/foundation-models Wed, 31 Dec 2025 12:00:00 +0000 Bradley J. Erickson AI Medical foundation models are typically restricted to one modality and one anatomical region. We review why that narrowness follows from the data, what data… https://mdsynapse.org/blogposts/algorithm-bias https://mdsynapse.org/blogposts/algorithm-bias Sat, 27 Dec 2025 12:00:00 +0000 Bradley J. Erickson Ethics A model with 92% accuracy in White patients and 78% in Black patients presents an adoption decision that aggregate accuracy cannot resolve. https://mdsynapse.org/blogposts/from-lab-to-ward https://mdsynapse.org/blogposts/from-lab-to-ward Sat, 27 Dec 2025 12:00:00 +0000 Bradley J. Erickson AI Medical AI systems that perform well in validation frequently fail to achieve sustained clinical use. https://mdsynapse.org/blogposts/model-drift https://mdsynapse.org/blogposts/model-drift Sat, 27 Dec 2025 12:00:00 +0000 Bradley J. Erickson AI Model drift is the progressive loss of accuracy that occurs when operational data diverge from a model's training distribution.