Breast cancer detection with IBM PowerAI Vision
summary
Proof‑of‑technology uses IBM AC922 Deep Learning System and IBM PowerAI Vision to classify histopathology slides (over 10,000 microscopy images) with a GoogleNet + class activation mapping model on 40X magnification, improving from 87% to 93% accuracy after automated augmentation (flips/rotations), providing heatmap explanations and sub‑second inference. Pipeline runs on‑prem for regulatory reasons, supports DICOM metadata extraction, model monitoring per FDA software‑as‑medical‑device lifecycle, and targets integration for combined scoring across mammography, MRI and histology.
AI-written summary · about
chapters (16)
- 0:00Introduction to breast cancer detection
- 0:27Problem statement and diagnostic stages
- 3:16Survival rates and urgency of early detection
- 4:41Accelerating diagnosis with deep learning
- 5:23Regulations for AI in medical devices
- 6:46Infrastructure and data handling challenges
- 8:54Histopathology and file format complexities
- 10:38Goal of AI-powered diagnostic pipeline
- 11:00Hardware and software setup with IBM PowerAI Vision
- 12:25Automating the diagnostic pipeline
- 13:29Experiment with histopathology images
- 15:58Results and accuracy improvements
- 16:41Extending to mammography and MRI
- 17:45Multi-disciplinary cancer case scoring
- 18:49Deploying models in production environments
- 19:53Q&A session
related talks
29mCancer diagnostics enabled by machines: finding and classifying single tumor....
19mCOVID Detection - Deep Learning Solution on power 9 AI Box
34mEnabling Early Stage Breast Cancer Detection Using AI
21mCreating Markets, Creating Trustworthiness: Standards for AI
01h 02mArtificial Intelligence for Early Detection and Diagnosis of Cervical Cancer
28mHow AI Can Help Detect Breast Cancer
55mDeep Learning in Medical Image Diagnostics
21mFighting Cancer with AI