Add MRI brain classification models (93.95% tumor accuracy)

README.md

@@ -0,0 +1,160 @@
+# MRI Brain Tumor Classification Models
+
+This repository contains trained deep learning models for MRI brain scan classification, developed using Mojo and PyTorch.
+
+## Models
+
+### 1. Brain Tumor 2D CNN (`kaggle_tumor_2dcnn_best.pth`)
+
+**Task:** Multi-class brain tumor classification from MRI slices
+
+| Metric | Value |
+|--------|-------|
+| **Accuracy** | 93.95% |
+| **Precision** | 0.94 |
+| **Recall** | 0.94 |
+| **F1 Score** | 0.94 |
+
+**Classes:**
+- `glioma` - 98.1% accuracy
+- `meningioma` - 83.9% accuracy
+- `notumor` - 98.5% accuracy
+- `pituitary` - 94.3% accuracy
+
+**Architecture:**
+- 2D CNN with 4 convolutional blocks
+- BatchNorm + Dropout regularization
+- AdaptiveAvgPool + 3-layer classifier
+- 4.85M parameters
+
+**Training:**
+- Dataset: Kaggle Brain Tumor MRI (7,023 images)
+- Input: 224x224 RGB images
+- Epochs: 50
+- Optimizer: AdamW with OneCycleLR
+- Augmentation: RandomCrop, Flip, Rotation, ColorJitter, GaussianBlur
+
+### 2. IXI 3D Brain CNN (`ixi_3dcnn_best.pth`)
+
+**Task:** 3D brain MRI volume classification
+
+**Architecture:**
+- 3D CNN with 4 Conv3D blocks
+- BatchNorm3D + Dropout3D
+- Global Average Pooling
+- 1.2M parameters
+
+**Training:**
+- Dataset: IXI Brain MRI (681 NIfTI volumes)
+- Input: 64x64x64 3D volumes
+- Epochs: 30
+- GPU: NVIDIA RTX 4090
+
+## Usage
+
+### Load Tumor Model (PyTorch)
+
+```python
+import torch
+import torch.nn as nn
+
+class TumorCNN(nn.Module):
+    def __init__(self, num_classes=4):
+        super(TumorCNN, self).__init__()
+        self.features = nn.Sequential(
+            nn.Conv2d(3, 64, 3, padding=1), nn.BatchNorm2d(64), nn.ReLU(),
+            nn.Conv2d(64, 64, 3, padding=1), nn.BatchNorm2d(64), nn.ReLU(),
+            nn.MaxPool2d(2, 2), nn.Dropout2d(0.25),
+            
+            nn.Conv2d(64, 128, 3, padding=1), nn.BatchNorm2d(128), nn.ReLU(),
+            nn.Conv2d(128, 128, 3, padding=1), nn.BatchNorm2d(128), nn.ReLU(),
+            nn.MaxPool2d(2, 2), nn.Dropout2d(0.25),
+            
+            nn.Conv2d(128, 256, 3, padding=1), nn.BatchNorm2d(256), nn.ReLU(),
+            nn.Conv2d(256, 256, 3, padding=1), nn.BatchNorm2d(256), nn.ReLU(),
+            nn.MaxPool2d(2, 2), nn.Dropout2d(0.25),
+            
+            nn.Conv2d(256, 512, 3, padding=1), nn.BatchNorm2d(512), nn.ReLU(),
+            nn.Conv2d(512, 512, 3, padding=1), nn.BatchNorm2d(512), nn.ReLU(),
+            nn.AdaptiveAvgPool2d((1, 1)),
+        )
+        self.classifier = nn.Sequential(
+            nn.Flatten(),
+            nn.Linear(512, 256), nn.ReLU(), nn.Dropout(0.5),
+            nn.Linear(256, 128), nn.ReLU(), nn.Dropout(0.3),
+            nn.Linear(128, num_classes),
+        )
+    
+    def forward(self, x):
+        return self.classifier(self.features(x))
+
+# Load model
+model = TumorCNN(num_classes=4)
+model.load_state_dict(torch.load("kaggle_tumor_2dcnn_best.pth"))
+model.eval()
+
+# Inference
+from torchvision import transforms
+from PIL import Image
+
+transform = transforms.Compose([
+    transforms.Resize((224, 224)),
+    transforms.ToTensor(),
+    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
+])
+
+image = Image.open("brain_mri.jpg").convert("RGB")
+input_tensor = transform(image).unsqueeze(0)
+
+with torch.no_grad():
+    output = model(input_tensor)
+    pred = output.argmax(1).item()
+
+classes = ['glioma', 'meningioma', 'notumor', 'pituitary']
+print(f"Prediction: {classes[pred]}")
+```
+
+## Training
+
+```bash
+# Clone repository
+git clone https://huggingface.co/YOUR_USERNAME/mri-brain-classification
+
+# Train tumor model
+mojo run scripts/train_tumor.mojo
+
+# Train 3D model
+mojo run scripts/train_advanced.mojo
+```
+
+## Evaluation
+
+```bash
+# Evaluate tumor model
+mojo run scripts/evaluate_tumor.mojo
+
+# Evaluate IXI model
+mojo run scripts/evaluate_ixi.mojo
+```
+
+## Citation
+
+```bibtex
+@misc{mri-brain-classification,
+  author = {Meidverse},
+  title = {MRI Brain Tumor Classification Models},
+  year = {2024},
+  publisher = {Hugging Face},
+  url = {https://huggingface.co/Meidverse/mri-brain-classification}
+}
+```
+
+## License
+
+MIT License
+
+## Acknowledgments
+
+- [Kaggle Brain Tumor MRI Dataset](https://www.kaggle.com/datasets/masoudnickparvar/brain-tumor-mri-dataset)
+- [IXI Dataset](https://brain-development.org/ixi-dataset/)
+- Built with Mojo 🔥 and PyTorch

best_model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de1927dcca7dbf627c90128a2610b0d44b0f7b78018a878c73616c870f7b0112
+size 4801205

config.json

@@ -0,0 +1,57 @@
+{
+  "tumor_model": {
+    "name": "kaggle_tumor_2dcnn_best",
+    "task": "image-classification",
+    "architecture": "2D-CNN",
+    "num_classes": 4,
+    "classes": ["glioma", "meningioma", "notumor", "pituitary"],
+    "input_size": [224, 224],
+    "channels": 3,
+    "parameters": 4853956,
+    "metrics": {
+      "accuracy": 0.9395,
+      "precision": 0.94,
+      "recall": 0.94,
+      "f1_score": 0.94
+    },
+    "per_class_accuracy": {
+      "glioma": 0.981,
+      "meningioma": 0.839,
+      "notumor": 0.985,
+      "pituitary": 0.943
+    },
+    "training": {
+      "dataset": "Kaggle Brain Tumor MRI",
+      "samples": 7023,
+      "epochs": 50,
+      "batch_size": 32,
+      "optimizer": "AdamW",
+      "learning_rate": 0.0001,
+      "scheduler": "OneCycleLR",
+      "gpu": "NVIDIA RTX 4090"
+    }
+  },
+  "ixi_model": {
+    "name": "ixi_3dcnn_best",
+    "task": "3d-volume-classification",
+    "architecture": "3D-CNN",
+    "num_classes": 2,
+    "classes": ["healthy", "diseased"],
+    "input_size": [64, 64, 64],
+    "channels": 1,
+    "parameters": 1196674,
+    "training": {
+      "dataset": "IXI Brain MRI",
+      "samples": 681,
+      "epochs": 30,
+      "batch_size": 2,
+      "optimizer": "Adam",
+      "learning_rate": 0.0001
+    }
+  },
+  "framework": {
+    "primary": "Mojo",
+    "inference": "PyTorch",
+    "version": "2.0+"
+  }
+}

ixi_3dcnn_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e3647054a8919ace03f3d9c15012b8aa9263888fe1022332b670de12e217307
+size 4801205

kaggle_tumor_2dcnn_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c20abe0c57f018dda98486712fdf4e47d5cb118e1aa5ee0d03109e927ec48218
+size 19452813