Deep Learning in Medical Image Diagnostics
summary
Convolutional CNN systems are applied to two pilots: a breast cancer pipeline that tiles whole‑slide images into patches, performs nuclei segmentation and mitotic count-based grading using transfer learning and fine‑tuned heads, and a diabetic retinopathy model trained from ~35,000 fundus photos (moved from CPU→GPU→NVIDIA DGX‑1) where keeping color and native resolution improves accuracy versus early grayscale normalization. Both are in pilot use, face label scarcity and class imbalance, and require explainability for clinical trust.
AI-written summary · about
chapters (37)
- 0:00Introduction
- 2:40Definition of Medical Imaging
- 3:35Types of Medical Imaging
- 5:06Computer Aided Diagnostics
- 6:40Medical Imaging Evolution
- 9:15Where are we now
- 10:25Disclaimer
- 12:13Academic Perspective
- 12:27Recent Publications
- 13:40Current Challenges
- 14:01Label Desert
- 14:22Data Overload
- 14:45Digital Silos
- 15:17Compliance in Governance
- 15:45Trust
- 16:35History of Deep Learning
- 19:10Feature Engineering
- 19:46Classification and Localization
- 20:14Object Deduction
- 20:32Segmentation
- 21:01Classic ML vs Deep Learning
- 21:55Manual Intervention
- 22:43Classical ML vs Deep Learning
- 24:08What is CNN
- 27:40Transfer Learning
- 29:29Data argumentation
- 29:55Commonly used techniques
- 31:07Domain Transfer Game
- 31:29Oneshot Learning
- 31:47Capsule Networks
- 32:18Experiences
- 32:49Breast Cancer
- 33:34Digital Whole Slide Images
- 36:05Diabetic Retinopathy
- 37:28Solution
- 40:25Learnings
- 42:41Questions
related talks
25mDeep Learning in Biomedicine: Challenges and Opportunities
29mCancer diagnostics enabled by machines: finding and classifying single tumor....
30mTransfer learning for image recognition in healthcare...
24mMachine Learning & Medical Imaging: The Future of Early Cancer Detection
01h 02mArtificial Intelligence for Early Detection and Diagnosis of Cervical Cancer
43mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
36mResponsible AI
33mUnsupervised and Semi-Supervised Deep Learning for Medical Imaging