Add MedMNIST tutorial for issue #172

[Zeesejo]

This notebook demonstrates how to use AUCMEDI with the MedMNIST benchmark datasets for medical image classification, including setup, data loading, and model training.

Closes #172

[Copilot]

Pull request overview

Adds a new tutorial notebook demonstrating how to use AUCMEDI with the MedMNIST benchmark (PathMNIST) for medical image classification (issue #172).

Changes:

• Introduces examples/tutorials/MedMNIST.ipynb covering installation, data loading, AUCMEDI pipeline setup, training, evaluation, and visualization.
• Demonstrates a DenseNet-based training workflow on PathMNIST.

Comments suppressed due to low confidence (1)

examples/tutorials/MedMNIST.ipynb:2

• This notebook file is committed as a minified one-line JSON blob (no stable formatting / cell IDs), whereas other notebooks in examples/tutorials/ are pretty-printed with one JSON field per line. The minified format makes reviews and future diffs very hard to read/maintain. Re-save the notebook with standard Jupyter formatting (and ideally clear outputs) before merging.

{"cells":[{"cell_type":"markdown","metadata":{},"source":["# Medical Image Classification with MedMNIST\n","\n","This notebook demonstrates how to use **AUCMEDI** with the **MedMNIST** benchmark datasets.\n","\n","MedMNIST is a collection of standardized biomedical image datasets for medical image analysis. We'll use **PathMNIST** (pathology images) to demonstrate AUCMEDI's three pillars: DataInterface, NeuralNetwork, and DataGenerator."]},{"cell_type":"markdown","metadata":{},"source":["## Setup and Installation\n","\n","First, install the medmnist package:"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["!pip install medmnist\n","import os\n","os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"0\""]},{"cell_type":"markdown","metadata":{},"source":["## Import Libraries"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["import numpy as np\n","from medmnist import PathMNIST\n","import aucmedi\n","from aucmedi import *"]},{"cell_type":"markdown","metadata":{},"source":["## Load MedMNIST Data\n","\n","We'll use PathMNIST, which contains 107,180 pathology images from 9 tissue classes."]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Download and load PathMNIST\n","train_dataset = PathMNIST(split='train', download=True)\n","val_dataset = PathMNIST(split='val', download=True)\n","test_dataset = PathMNIST(split='test', download=True)\n","\n","print(f'Train: {len(train_dataset)} samples')\n","print(f'Val: {len(val_dataset)} samples')\n","print(f'Test: {len(test_dataset)} samples')"]},{"cell_type":"markdown","metadata":{},"source":["## Convert to AUCMEDI Format\n","\n","AUCMEDI expects data in a specific format. We'll convert the MedMNIST data:"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Prepare data for AUCMEDI\n","# Create sample lists and labels\n","train_samples = []\n","train_labels = []\n","for i in range(len(train_dataset)):\n","    train_samples.append(train_dataset[i][0])\n","    train_labels.append(train_dataset[i][1])\n","\n","test_samples = []\n","test_labels = []\n","for i in range(len(test_dataset)):\n","    test_samples.append(test_dataset[i][0])\n","    test_labels.append(test_dataset[i][1])\n","\n","# Convert to numpy arrays\n","train_x = np.array([np.array(img) for img in train_samples])\n","train_y = np.array(train_labels).flatten()\n","test_x = np.array([np.array(img) for img in test_samples])\n","test_y = np.array(test_labels).flatten()\n","\n","print(f'Train images shape: {train_x.shape}')\n","print(f'Train labels shape: {train_y.shape}')\n","print(f'Number of classes: {len(np.unique(train_y))}')"]},{"cell_type":"markdown","metadata":{},"source":["## AUCMEDI's Three Pillars\n","\n","AUCMEDI is built on three pillars:\n","1. **DataInterface**: Handles data loading\n","2. **NeuralNetwork**: Manages the model architecture\n","3. **DataGenerator**: Provides data augmentation and preprocessing"]},{"cell_type":"markdown","metadata":{},"source":["### 1. DataInterface Setup"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Create index lists\n","train_samples_idx = list(range(len(train_x)))\n","test_samples_idx = list(range(len(test_x)))\n","\n","# Initialize DataInterface for in-memory data\n","ds = DataInterface(interface=\"numpy\",\n","                   data_directory=None,\n","                   training_samples=train_samples_idx,\n","                   validation_samples=None,\n","                   test_samples=test_samples_idx)\n","\n","print('DataInterface initialized')"]},{"cell_type":"markdown","metadata":{},"source":["### 2. NeuralNetwork Setup\n","\n","We'll use a DenseNet121 architecture pretrained on ImageNet:"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Define model\n","model = NeuralNetwork(n_labels=9,\n","                      channels=3,\n","                      architecture=\"DenseNet121\",\n","                      pretrained_weights=True,\n","                      loss=\"categorical_crossentropy\",\n","                      metrics=[\"accuracy\"])\n","\n","print('Model architecture:', model.architecture)\n","print('Input shape:', (28, 28, 3))"]},{"cell_type":"markdown","metadata":{},"source":["### 3. DataGenerator Setup\n","\n","Configure data augmentation and preprocessing:"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["from aucmedi.data_processing.subfunctions import Resize\n","\n","# Create subfunctions for preprocessing\n","# Resize images to 224x224 for DenseNet\n","sf_list = [Resize(shape=(224, 224))]\n","\n","# Initialize DataGenerators\n","train_gen = DataGenerator(train_samples_idx,\n","                          path_imagedir=None,\n","                          labels=train_y,\n","                          image_format=\"array\",\n","                          batch_size=32,\n","                          data_aug=None,\n","                          shuffle=True,\n","                          subfunctions=sf_list,\n","                          resize=None,\n","                          standardize_mode=\"tf\",\n","                          grayscale=False,\n","                          sample_weights=None,\n","                          seed=None,\n","                          image_loader=lambda idx: train_x[idx])\n","\n","test_gen = DataGenerator(test_samples_idx,\n","                         path_imagedir=None,\n","                         labels=test_y,\n","                         image_format=\"array\",\n","                         batch_size=32,\n","                         data_aug=None,\n","                         shuffle=False,\n","                         subfunctions=sf_list,\n","                         resize=None,\n","                         standardize_mode=\"tf\",\n","                         grayscale=False,\n","                         sample_weights=None,\n","                         seed=None,\n","                         image_loader=lambda idx: test_x[idx])\n","\n","print('DataGenerators initialized')"]},{"cell_type":"markdown","metadata":{},"source":["## Training\n","\n","Train the model on PathMNIST data:"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Train model\n","history = model.train(train_gen,\n","                      epochs=10,\n","                      validation_freq=1,\n","                      callbacks=[])\n","\n","print('Training complete!')"]},{"cell_type":"markdown","metadata":{},"source":["## Evaluation\n","\n","Evaluate the model on the test set:"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["# Predict on test set\n","preds = model.predict(test_gen)\n","\n","# Calculate accuracy\n","from sklearn.metrics import accuracy_score, classification_report\n","\n","pred_labels = np.argmax(preds, axis=1)\n","accuracy = accuracy_score(test_y, pred_labels)\n","\n","print(f'Test Accuracy: {accuracy:.4f}')\n","print('\\nClassification Report:')\n","print(classification_report(test_y, pred_labels))"]},{"cell_type":"markdown","metadata":{},"source":["## Visualization\n","\n","Visualize some predictions:"]},{"cell_type":"code","execution_count":null,"metadata":{},"outputs":[],"source":["import matplotlib.pyplot as plt\n","\n","# Show sample predictions\n","fig, axes = plt.subplots(2, 5, figsize=(15, 6))\n","axes = axes.ravel()\n","\n","class_names = ['ADI', 'BACK', 'DEB', 'LYM', 'MUC', 'MUS', 'NORM', 'STR', 'TUM']\n","\n","for i in range(10):\n","    axes[i].imshow(test_x[i])\n","    axes[i].set_title(f'True: {class_names[test_y[i]]}\\nPred: {class_names[pred_labels[i]]}')\n","    axes[i].axis('off')\n","\n","plt.tight_layout()\n","plt.show()"]},{"cell_type":"markdown","metadata":{},"source":["## Summary\n","\n","This tutorial demonstrated:\n","- Loading MedMNIST datasets\n","- Converting to AUCMEDI format\n","- Using AUCMEDI's three pillars (DataInterface, NeuralNetwork, DataGenerator)\n","- Training and evaluating on medical imaging data\n","\n","Try experimenting with:\n","- Different MedMNIST datasets (ChestMNIST, DermaMNIST, etc.)\n","- Different architectures (ResNet, EfficientNet, etc.)\n","- Data augmentation techniques\n","- Ensemble methods"]}],"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"},"language_info":{"name":"python","version":"3.8.0"}},"nbformat":4,"nbformat_minor":4}

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.