New trials compare Claude, ChatGPT, and DeepSeek on mammography reports


 While AI is excelling at administrative tasks it still struggles to beat the human eye in complex cancer diagnosis. A comparative study pitted three major language models—ChatGPT-4o, Claude 3 Opus, and DeepSeek-R1—against human radiologists in analyzing mammography reports for breast cancer risks (BI-RADS 4). The results were clear: human radiologists significantly outperformed all three AI models. While the AI tools demonstrated high sensitivity meaning they were good at flagging potential issues they suffered from low specificity leading to a high volume of false alarms. The study concludes that for now these tools are best used as "safety net" assistants to ensure nothing is missed rather than as independent diagnostic agents.

Read the original article at: https://medinform.jmir.org/2025/1/e80182


Follow us on Instagram, Twitter, and Facebook to stay up to date with what's new in healthcare all around the world.

Comments

Post a Comment

Popular posts from this blog

Cultural barriers and privacy fears are stalling digital adoption

Image
  Despite the rapid proliferation of mobile health (mHealth) tools designed to manage chronic diseases and promote wellness, a significant adoption gap persists among racially diverse and underserved populations. A recent review indicates that while mHealth holds the promise of democratizing healthcare access, its benefits are unevenly distributed. The data suggests that the "digital divide" is no longer just about access to hardware; it is increasingly defined by a lack of trust. Many minority groups express deep reservations regarding data privacy and surveillance, fearing how their health information might be used or monetized by technology vendors. Beyond trust issues, the report identifies a lack of cultural competence in app design as a primary barrier to entry. Many existing digital health solutions fail to account for linguistic nuances, health literacy levels, or specific cultural health beliefs, rendering them ineffective for the populations that need them most. To ...
Read more

Digital Health Insights: December 4th – 10th, 2025

Image
Sehatmandi app connects 189 remote facilities for real-time care A new qualitative study published in the Journal of Medical Internet Research highlights the operational impact of the Sehatmandi digital health intervention in Afghanistan. Designed to bridge the gap in healthcare delivery across conflict-affected regions, the app has successfully connected 189 diverse health facilities, ranging from comprehensive health centers to smaller sub-centers. The platform enables real-time data submission, allowing for immediate monitoring of pharmaceutical stocks, patient service delivery, and staff attendance. This transition from paper-based reporting to digital logs has significantly improved transparency and accountability in a notoriously difficult operating environment. However, the implementation has not been without significant structural hurdles. The study reveals that while the app streamlined data flow, frontline workers faced persistent challenges including unstable internet conne...
Read more

Supercomputers reveal a new Parkinson's culprit: malfunctioning PT5B neurons that trigger the chaotic brain waves behind tremors

Image
 In a breakthrough facilitated by high-performance computing, researchers have identified a specific neuron type—"PT5B" neurons in the motor cortex—as a primary driver of Parkinson's disease symptoms. By using supercomputers to simulate the brain's complex electrical circuitry, scientists were able to isolate how these specific neurons malfunction. The simulations revealed that when PT5B neurons fall into a "hypersynchronized" state, they generate the chaotic beta oscillations (brain waves) responsible for the tremors and rigidity characteristic of the disease. This finding is significant because it moves beyond the general understanding of dopamine loss to a precise circuit-level mechanism. It offers a new, highly specific target for future therapies. Instead of broad pharmaceutical approaches that often carry heavy side effects, future treatments could theoretically target the stabilization of PT5B neurons specifically, potentially silencing the "no...
Read more