Home

Medical-AI is an AI framework for rapid prototyping/experimentation of AI for Medical Applications

Documentation: https://aibharata.github.io/medicalAI/

Source Code: https://github.com/aibharata/medicalai

Youtube Tutorial:

Medical-AI is an AI framework for rapid prototyping of AI for Medical Applications.

Installation¶

$ pip install medicalai

---> 100%

Requirements¶

Python Version : 3.5-3.7 (Doesn't Work on 3.8 Since Tensorflow does not support 3.8 yet.

Dependencies: Numpy, Tensorflow, Seaborn, Matplotlib, Pandas

NOTE: Dependency libraries are automatically installed. No need for user to install them manually.

Usage¶

Getting Started Tutorial: Google Colab¶

Google Colab Notebook Link

Importing the Library¶

import medicalai as ai

Using Templates¶

You can use the following templates to perform specific Tasks

Load Dataset From Folder¶

Set the path of the dataset and set the target dimension of image that will be input to AI network.

trainSet,testSet,labelNames =ai.datasetFromFolder(datasetFolderPath, targetDim = (96,96)).load_dataset()

- trainSet contains 'data' and 'labels' accessible by trainSet.data and trainSet.labels - testSet contains 'data' and 'labels' accessible by testSet.data and testSet.labels - labelNames contains class names/labels

Check Loaded Dataset Size¶

print(trainSet.data.shape)
print(trainSet.labels.shape)

Run Training and Save Model¶

trainer = ai.TRAIN_ENGINE()
trainer.train_and_save_model(AI_NAME= 'tinyMedNet', MODEL_SAVE_NAME='PATH_WHERE_MODEL_IS_SAVED_TO', trainSet, testSet, OUTPUT_CLASSES, RETRAIN_MODEL= True, BATCH_SIZE= 32, EPOCHS= 10, LEARNING_RATE= 0.001)

Plot Training Loss and Accuracy¶

trainer.plot_train_acc_loss()

Generate a comprehensive evaluation PDF report¶

trainer.generate_evaluation_report()

PDF report will be generated with model sensitivity, specificity, accuracy, confidence intervals, ROC Curve Plot, Precision Recall Curve Plot, and Confusion Matrix Plot for each class. This function can be used when evaluating a model with Test or Validation Data Set.

Explain the Model on a sample¶

trainer.explain(testSet.data[0:1], layer_to_explain='CNN3')

Loading Model for Prediction¶

infEngine = ai.INFERENCE_ENGINE(modelName = 'PATH_WHERE_MODEL_IS_SAVED_TO')

Predict With Labels¶

infEngine.predict_with_labels(testSet.data[0:2], top_preds=3)

Get Just Values of Prediction without postprocessing¶

infEngine.predict(testSet.data[0:2])

Alternatively, use a faster prediction method in production¶

infEngine.predict_pipeline(testSet.data[0:1])

Advanced Usage¶

Code snippet for Training Using Medical-AI¶

## Setup AI Model Manager with required AI. 
model = ai.modelManager(AI_NAME= AI_NAME, modelName = MODEL_SAVE_NAME, x_train = train_data, OUTPUT_CLASSES = OUTPUT_CLASSES, RETRAIN_MODEL= RETRAIN_MODEL)

# Start Training
result = ai.train(model, train_data, train_labels, BATCH_SIZE, EPOCHS, LEARNING_RATE, validation_data=(test_data, test_labels), callbacks=['tensorboard'])

# Evaluate Trained Model on Test Data
model.evaluate(test_data, test_labels)

# Plot Accuracy vs Loss for Training
ai.plot_training_metrics(result)

#Save the Trained Model
ai.save_model_and_weights(model, outputName= MODEL_SAVE_NAME)

Automated Tests¶

To Check the tests

    pytest

To See Output of Print Statements

    pytest -s