Update README.md
Browse files
README.md
CHANGED
|
@@ -8,7 +8,7 @@ tags:
|
|
| 8 |
- 3d-resnet
|
| 9 |
- monai
|
| 10 |
- oncology
|
| 11 |
-
- breast-
|
| 12 |
- sustainable-ai
|
| 13 |
- explainable-ai
|
| 14 |
metrics:
|
|
@@ -23,7 +23,7 @@ This repository contains the "Better Model" (Phase 3) for breast cancer classifi
|
|
| 23 |
## Key Engineering Breakthroughs (Phase 3)
|
| 24 |
- **Instance Normalization Integration:** Replaced Batch Normalization with **InstanceNorm 3D**. This critical fix allows the model to maintain stable gradients even with a Batch Size of 1, preventing the collapse observed in earlier versions.
|
| 25 |
- **Dynamic Temporal Fusion:** Uses a Multi-Stream approach where the Kinetic stream processes a variable number of post-contrast phases.
|
| 26 |
-
- **Improved Temporal Aggregation:**
|
| 27 |
- **Explainable AI (XAI):** Full support for **3D Grad-CAM** visualization on the T2 structural stream.
|
| 28 |
-
|
| 29 |
## Mathematical & Architectural Refinement
|
|
|
|
| 8 |
- 3d-resnet
|
| 9 |
- monai
|
| 10 |
- oncology
|
| 11 |
+
- breast-cancera
|
| 12 |
- sustainable-ai
|
| 13 |
- explainable-ai
|
| 14 |
metrics:
|
|
|
|
| 23 |
## Key Engineering Breakthroughs (Phase 3)
|
| 24 |
- **Instance Normalization Integration:** Replaced Batch Normalization with **InstanceNorm 3D**. This critical fix allows the model to maintain stable gradients even with a Batch Size of 1, preventing the collapse observed in earlier versions.
|
| 25 |
- **Dynamic Temporal Fusion:** Uses a Multi-Stream approach where the Kinetic stream processes a variable number of post-contrast phases.
|
| 26 |
+
- **Improved Temporal Aggregation:** Selected Temporal Mean Pooling over Recurrent units (LSTM/RNN) to ensure a more robust global temporal summary and prevent overfitting on this specific dataset size.
|
| 27 |
- **Explainable AI (XAI):** Full support for **3D Grad-CAM** visualization on the T2 structural stream.
|
| 28 |
-
|
| 29 |
## Mathematical & Architectural Refinement
|