Deep Learning in Biomedicine: Challenges and Opportunities
summary
Describes concrete results: a skin lesion model trained on ~130,000 clinical photos and pretrained on 1.3M nonmedical images matches or exceeds dermatologist accuracy; a lung cancer prognosis model uses CT volumes from Stanford, Florida and Maastricht cohorts and shows 3D convolutional networks outperform 2D and that combining images with clinical variables separates high‑ and low‑risk survival groups better than clinical data alone. Also presents a TensorFlow brain tumor segmentation pipeline on the BraTS set (~220 cases, 11‑reader labels). Notes privacy, dataset size and cross‑site bias as the main hurdles.
AI-written summary · about
chapters (28)
- 0:00Intro
- 2:14Precisions medicine
- 2:47Big data vs. data science
- 3:50Precision medicine
- 4:48Biomedical data
- 6:13Fast technological development
- 6:57The Cancer Genome Atlas (TCGA)
- 7:56Digital pathology
- 8:23Biomedical images
- 9:23Introduction: Radiomics
- 9:51Introduction: Traditional Radiomics
- 10:07Semantic Features: Edge shape
- 10:25Computational Features: Overview
- 11:07Stanford Data Set
- 11:26Decision tree modeling
- 11:41Classifier: Decision trees
- 11:58Introduction: Deep learning
- 12:26Deep learning for medicine
- 13:33Convolutional Neural Network
- 14:30Melanoma diagnosis
- 15:21Different types of images
- 17:23Lessons learned
- 18:06Background and Motivation
- 18:36Conventional Approach
- 22:19Brain Tumor Segmentation (Brats) Challenge
- 22:48Ground truth
- 23:16Visualizations
- 23:45Overall summary
related talks
55mDeep Learning in Medical Image Diagnostics
37mAI in medicine: where do we stand?
29mCancer diagnostics enabled by machines: finding and classifying single tumor....
41mDeep Learning to Solve Challenging Problems
15mBrain Tumor Detection and Classification Using Deep Learning Models
43mWhy you can't afford to ignore deep learning
24mMachine Learning & Medical Imaging: The Future of Early Cancer Detection
36mResponsible AI