Microsoft debuts an advanced diagnostic AI model, while major studies highlight both technical progress and patient trust concerns in medical AI.
Key Details
- 1Microsoft's MAI-DxO AI outperformed physicians on 85% of 304 NEJM real-world cases, reportedly quadrupling expert diagnostic rates.
- 2MIT and Mass General Brigham announced a new Analog Devices-funded program for joint medical AI research, aiming for six projects yearly.
- 3MIT researchers found LLMs may misdirect patients using colloquial symptom descriptions, with particular risk for female patients.
- 4A survey for ModMed found 55% of patients are uncomfortable with AI in diagnosis/treatment, yet 57% support it if it improves doctor-patient face time.
- 5New studies confirm patients perceive less empathy from AI-generated responses than from humans, impacting trust.
- 6Recent imaging AI highlights include tools for lung tumor segmentation, dementia diagnosis, and fatty liver detection via radiographs.
Why It Matters
Microsoft's LLM performance signals a leap in AI diagnostic accuracy and resource management, while the research on communication and trust underscores persistent challenges for clinical AI adoption. The breadth of new imaging AI applications and landmark academic partnerships indicate accelerating innovation but also highlight the need to address real-world usability and acceptance.
Source
AI in Healthcare
Related News
•Health Imaging
BMI Significantly Impacts AI Accuracy in CT Lung Nodule Detection
New research demonstrates that high BMI negatively impacts both human and AI performance in chest low-dose CT interpretation, highlighting dataset diversity concerns.
•Health Imaging
Most FDA-Cleared AI Devices Lack Pre-Approval Safety Data, Study Finds
A new study finds fewer than 30% of FDA-cleared AI medical devices reported key safety or adverse event data before approval.
•Radiology Business
Cigna Expands Nationwide Coverage for CT Imaging AI Tools
Cigna will reimburse for CT-based plaque analysis AI software for millions of members nationwide.