Cancer diagnostics enabled by machines: finding and classifying single tumor....
summary
Claims that deep learning runs clinical breast cancer diagnostics at scale: a cascade of segmentation and high‑magnification cell‑classification models processes gigapixel slides sliced into patches, uses dot annotations for cell counts, and combines supervised and unsupervised training. Validation shows parity with pathologists via clinical studies and CE labeling. Production stack uses PyTorch distributed data‑parallel training, DVC for data reproducibility, and Nvidia docker images to speed iteration; deployment already serves 20,000 patients diagnosed across 17 labs in 6 countries.
AI-written summary · about
chapters (9)
- 0:00AI in cancer diagnostics
- 3:21Pathology workflow challenges
- 7:13Tumor detection models
- 10:45Accelerating AI in pathology
- 16:01Code quality for faster AI
- 19:30CI/CD for deep learning
- 22:19Multi-GPU model training
- 26:34Learning from AI failures
- 28:41Key takeaways
related talks
55mDeep Learning in Medical Image Diagnostics
24mBreast cancer detection with IBM PowerAI Vision
25mDeep Learning in Biomedicine: Challenges and Opportunities
11mDeep learning helping to solve cancer
01h 02mArtificial Intelligence for Early Detection and Diagnosis of Cervical Cancer
18mCancer Screening using Deep Learning
18mCan AI make us smarter than cancer
24mMachine Learning & Medical Imaging: The Future of Early Cancer Detection