DeepSeek: A New Player in AI for Healthcare

The new open-source LLM, DeepSeek, is creating buzz for its potential to transform AI in medicine and healthcare. Designed for transparency and collaboration, it opens doors for innovation in medical research, decision support, and patient care. It's exciting to see what DeepSeek will bring to these conversations—I’ll keep you posted!

Deep learning model halves lung tumor segmentation times

Researchers at Stanford University have developed a cutting-edge deep learning model that halves the time needed to segment lung tumors on CT scans. Trained on one of the largest datasets of its kind—1,504 scans with 1,828 segmented tumors—the model uses an advanced 3D U-Net architecture to deliver near-expert-level performance.

Why It Matters

This model achieves 92% sensitivity and 82% specificity, with a Dice similarity coefficient (DSC) nearly matching that of radiologists. By reducing segmentation time from an average of 166 seconds to just 76 seconds, the tool enhances efficiency and supports faster treatment planning for lung cancer patients.

What’s Next?

With its ability to analyze scans from diverse CT equipment and detect smaller lesions, this innovation is poised to revolutionize lung cancer diagnostics and management.

​

TotalSegmentator v2.5

Total Segmentator folks have done it again! A new update to this amazing tool has been released. Here's what has been added:

MR:

• Individual vertebrae
• Appendicular bones
• Body trunk and extremities
• Thigh and shoulder muscles (also for CT)

CT:

• Intervertebral discs
• Intermuscular fat
• Lung nodules
• Breasts
• Oculomotor muscles
• Improved coronary arteries
• Improved kidney cysts

You can explore TotalSegmentator directly on their github repo here.