Neural Networks and Brains
summary
Uses TensorFlow with 3D convolutional nets (Xavier init, Adam optimizer, L2 + dropout) to segment glioblastoma from four MRI modalities. Model trains on BRATS high‑grade set of 220 brains, uses 25x25x25x4 patches with central 9x9x9 outputs, samples 50% of patches from tumor to fix class imbalance, and applies 1x1x1 convs with 50% dropout. Resulting average whole‑tumor Dice 0.86 (vs inter‑rater ~0.88), works well on solid tumors but shows higher variance and worse performance on diffuse lesions.
AI-written summary · about
chapters (15)
- 0:00Start
- 0:21Why Deep Learning Is Important in the Medical Domain
- 8:36Segmentation Task for Brain Cancer
- 9:19Outcome Prediction for Lung Cancer
- 12:00Training with Small Batch Size
- 12:47Example of a 3d Patch Based Architecture for a Segmentation Task
- 14:09Splice Based Architecture
- 16:31Slide Overview of Tensorflow
- 18:25Model Parameters
- 20:58Note on Tensorflow Scoping Mechanisms
- 25:53Glioblastoma Segmentation
- 30:19Slice Based Architecture
- 33:26Pre-Processing on the Brains
- 40:02Final Overview of the Network
- 41:10Results
related talks
15mBrain Tumor Detection and Classification Using Deep Learning Models
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
25mBig Brains, Small Targets: Whole-Brain Image Analysis with Python
47mThe whole truth about using deep learning in medical imaging
42mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
43mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
34mWhy AI in the Medical Field Both Sucks and Rocks...
01h 20mMRI Brain Scan Classification Guide | FastSurfer, XGBoost, 3D CNNs