medicalai.chief package¶

Subpackages¶

medicalai.chief.model_metrics package
- Submodules
- medicalai.chief.model_metrics.modelstats module
- Module contents
medicalai.chief.nnets package
- Submodules
- medicalai.chief.nnets.covid_net module
- medicalai.chief.nnets.densenet module
- medicalai.chief.nnets.inceptionResnet module
- medicalai.chief.nnets.inceptionv3 module
- medicalai.chief.nnets.mobilenet module
- medicalai.chief.nnets.mobilenetv2 module
- medicalai.chief.nnets.resnet module
- medicalai.chief.nnets.vgg16 module
- medicalai.chief.nnets.xception module
- Module contents
medicalai.chief.xai package
- Submodules
- medicalai.chief.xai.xcams module
- Module contents

Submodules¶

medicalai.chief.core module¶

class medicalai.chief.core.INFERENCE_ENGINE(modelName, testSet=None, classNames=None)¶

Bases: object

TODO: Need to add Metaloader support

decode_predictions(pred, top_preds=4, retType='tuple')¶

explain(input, predictions=None, layer_to_explain='CNN3', classNames=None, selectedClasses=None, expectedClass=None, showPlot=False)¶

generate_evaluation_report(testSet=None, predictions=None, printStat=False, returnPlot=False, showPlot=False, pdfName=None, **kwargs)¶

getLayerNames()¶

load_model_and_weights(modelName)¶

load_network(fileName)¶

load_weights(wgtfileName)¶

predict(input)¶

predict_pipeline(input)¶

Slightly Faster version of predict. Useful for deployment.

preprocessor_from_meta(metaFile=None)¶

summary()¶

class medicalai.chief.core.TRAIN_ENGINE(modelName=None)¶

Bases: medicalai.chief.core.INFERENCE_ENGINE

plot_train_acc_loss()¶

train_and_save_model(AI_NAME, MODEL_SAVE_NAME, trainSet, testSet, OUTPUT_CLASSES, RETRAIN_MODEL, BATCH_SIZE, EPOCHS, LEARNING_RATE, convLayers=None, SAVE_BEST_MODEL=True, BEST_MODEL_COND=None, callbacks=None, loss='sparse_categorical_crossentropy', metrics=['accuracy'], showModel=False, CLASS_WEIGHTS=None)¶

” CLASS_WEIGHTS: Dictionary containing class weights for model.fit()

medicalai.chief.core.check_model_exists(outputName)¶

medicalai.chief.core.create_model_output_folder(outputName)¶

medicalai.chief.core.decode_predictions(pred, labelNames, top_preds=4, retType='tuple')¶

medicalai.chief.core.load_model_and_weights(modelName, summary=False)¶

medicalai.chief.core.modelManager(modelName, x_train, OUTPUT_CLASSES, RETRAIN_MODEL, AI_NAME='tinyMedNet', convLayers=None)¶

medicalai.chief.core.plot_training_metrics(result, theme='light')¶

medicalai.chief.core.predict_labels(model, input, expected_output=None, labelNames=None, top_preds=4)¶

predict(x, batch_size=None, verbose=0, steps=None, callbacks=None, max_queue_size=10, workers=1, use_multiprocessing=False)

medicalai.chief.core.save_model_and_weights(model, outputName)¶

medicalai.chief.core.show_model_details(model)¶

medicalai.chief.core.train(model, x_train, batch_size=1, epochs=1, learning_rate=0.001, callbacks=None, class_weights=None, saveBestModel=False, bestModelCond=None, validation_data=None, TRAIN_STEPS=None, TEST_STEPS=None, loss='sparse_categorical_crossentropy', metrics=['accuracy'], verbose=None, y_train=None)¶

medicalai.chief.dataset_prepare module¶

class medicalai.chief.dataset_prepare.AUGMENTATION(rotation_range=12, fill_mode='constant', width_shift_range=0.1, height_shift_range=0.1, horizontal_flip=False, vertical_flip=False, brightness_range=(0.9, 1.1), zoom_range=(0.85, 1.15), rescale=0.00392156862745098, shear_range=0, channel_shift_range=0, samplewise_center=False, samplewise_std_normalization=False, featurewise_center=False, featurewise_std_normalization=False, cval=0, preprocessing_function=None)¶

Bases: object

create_aug()¶

class medicalai.chief.dataset_prepare.INPUT_PROCESSOR(targetDim=(31, 31), samplingMethod=None, normalize=False, color_mode='RGB', rescale=None, dtype='float32')¶

Bases: object

processImage(image)¶

resizeDataSet(dataset)¶

resizeDataSetfromFolder(folder)¶

class medicalai.chief.dataset_prepare.InputProcessorFromMeta(metaFile)¶

Bases: medicalai.chief.dataset_prepare.INPUT_PROCESSOR

medicalai.chief.dataset_prepare.convertlist2tuple(lst)¶

medicalai.chief.dataset_prepare.datasetFolderStructureValidate(folder)¶

class medicalai.chief.dataset_prepare.datasetFromFolder(folder, targetDim=(31, 31), normalize=False, name=None, useCache=True, forceCleanCache=False)¶

Bases: medicalai.chief.dataset_prepare.datasetManager

TODO: Fix samplingMethodName assignment

load_dataset()¶

class medicalai.chief.dataset_prepare.datasetGenFromDataframe(folder, csv_path='.', x_col='name', y_col='labels', targetDim=(224, 224), normalize=False, batch_size=16, augmentation=True, color_mode='rgb', class_mode='sparse', shuffle=True, seed=17)¶

Bases: object

Creates Keras Dataset Generator for Handling Large Datasets from DataFrame.

Parameters
- csv_path – folder containing train.csv and test.csv.
- folder – The directory must be set to the path where your training images are present.
- x_col – Name of column containing image name, default = name.
- y_col – Name of column for labels, default = labels.
- targetDim – The target_size is the size of your input images to the neural network.
- class_mode – Set binary if classifying only two classes, if not set to categorical, in case of an Autoencoder system, both input and the output would probably be the same image, for this case set to input.
- color_mode – grayscale for black and white or grayscale, rgb for three color channels.
- batch_size – Number of images to be yielded from the generator per batch. If training fails lower this number.

get_class_weights()¶

get_numpy(generator)¶

load_generator()¶

class medicalai.chief.dataset_prepare.datasetGenFromFolder(folder, targetDim=(224, 224), normalize=False, batch_size=16, augmentation=True, color_mode='rgb', class_mode='sparse', shuffle=True, seed=17)¶

Bases: object

folder : The directory must be set to the path where your n classes of folders are present. targetDim : The target_size is the size of your input images to the neural network. class_mode : Set binary if classifying only two classes, if not set to categorical, in case of an Autoencoder system, both input and the output would probably be the same image, for this case set to input. color_mode: grayscale for black and white or grayscale, rgb for three color channels. batch_size: Number of images to be yielded from the generator per batch. If training fails lower this number.

get_class_weights()¶

get_numpy(generator)¶

load_generator()¶

class medicalai.chief.dataset_prepare.datasetManager(folder, targetDim=(31, 31), normalize=False, name=None, useCache=True, forceCleanCache=False)¶

Bases: medicalai.chief.dataset_prepare.INPUT_PROCESSOR

compress_and_cache_data(**kw)¶

convert_dataset(**kw)¶

load_data()¶

process_dataset()¶

reload_data(**kw)¶

medicalai.chief.dataset_prepare.datasetManagerFunc(folder, targetDim=(31, 31), normalize=False)¶

medicalai.chief.dataset_prepare.getLabelsFromFolder(folder)¶

class medicalai.chief.dataset_prepare.medicalai_generator()¶

Bases: tensorflow.python.keras.preprocessing.image.ImageDataGenerator

medicalai.chief.dataset_prepare.metaLoader(metaFile)¶

medicalai.chief.dataset_prepare.metaSaver(labelMap, labels, normalize=None, rescale=None, network_input_dim=None, samplingMethodName=None, outputName=None)¶

class medicalai.chief.dataset_prepare.myDict()¶

Bases: dict

medicalai.chief.dataset_prepare.safe_labelmap_converter(labelMap)¶

medicalai.chief.download_utils module¶

class medicalai.chief.download_utils.DLProgress(iterable=None, desc=None, total=None, leave=True, file=None, ncols=None, mininterval=0.1, maxinterval=10.0, miniters=None, ascii=None, disable=False, unit='it', unit_scale=False, dynamic_ncols=False, smoothing=0.3, bar_format=None, initial=0, position=None, postfix=None, unit_divisor=1000, write_bytes=None, gui=False, **kwargs)¶

Bases: tqdm._tqdm.tqdm

hook(block_num=1, block_size=1, total_size=None)¶

last_block( = 0)¶

medicalai.chief.download_utils.check_if_url(x)¶

medicalai.chief.download_utils.getFile(url, storePath=None, cacheDir=None, subDir='dataset')¶

medicalai.chief.download_utils.load_image(link, target_size=(32, 32), storePath=None, cacheDir=None, subDir='images')¶

medicalai.chief.download_utils.untar(tar_file, destination)¶

medicalai.chief.download_utils.unzip(zip_file, destination)¶

medicalai.chief.networks module¶

class medicalai.chief.networks.DenseNet121()¶

Bases: medicalai.chief.networks.NetworkInit

DenseNet121 model, with weights pre-trained on ImageNet inputSize: input image size tuple outputSize: Number of classes for prediction

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.InceptionResNetV2()¶

Bases: medicalai.chief.networks.NetworkInit

InceptionResNetV2 model, with weights pre-trained on ImageNet inputSize: input image size tuple,default : (224,223,3) outputSize: Number of classes for prediction

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.InceptionV3()¶

Bases: medicalai.chief.networks.NetworkInit

InceptionV3 model, with weights pre-trained on ImageNet inputSize: input image size tuple,default : (224,223,3) outputSize: Number of classes for prediction

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.MobileNet()¶

Bases: medicalai.chief.networks.NetworkInit

MobileNet model, with weights pre-trained on ImageNet inputSize: input image size tuple,default : (224,223,3) outputSize: Number of classes for prediction

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.MobileNetV2()¶

Bases: medicalai.chief.networks.NetworkInit

MobileNet model, with weights pre-trained on ImageNet inputSize: input image size tuple,default : (224,223,3) outputSize: Number of classes for prediction

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.NetworkInit()¶

Bases: object

Base class for parameter Network initializers.

The NetworkInit class represents a network initializer used to initialize network/model parameters for numerous medical ai networks. It should be subclassed when implementing new types of network initializers.

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.VGG16()¶

Bases: medicalai.chief.networks.NetworkInit

VGG16 model, with weights pre-trained on ImageNet inputSize: input image size tuple,default : (224,223,3) outputSize: Number of classes for prediction

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.Xception()¶

Bases: medicalai.chief.networks.NetworkInit

Xception model, with weights pre-trained on ImageNet inputSize: input image size tuple,default : (224,223,3) outputSize: Number of classes for prediction

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

medicalai.chief.networks.get(networkInitialization)¶

class medicalai.chief.networks.megaNet()¶

Bases: medicalai.chief.networks.NetworkInit

megaNet is based on COVID-NET. This is a tensorflow 2.0 network variant for COVID-Net described in Paper “COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest Radiography Images” by Linda Wang et al. Reference: https://github.com/busyyang/COVID-19/

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.resNet110()¶

Bases: medicalai.chief.networks.NetworkInit

resnet110

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.resNet20()¶

Bases: medicalai.chief.networks.NetworkInit

resnet20

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.resNet32()¶

Bases: medicalai.chief.networks.NetworkInit

resnet32

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.resNet56()¶

Bases: medicalai.chief.networks.NetworkInit

RESNET56

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.tinyMedNet()¶

Bases: medicalai.chief.networks.NetworkInit

tinyMedNet is a classification network that consumes very less resources and can be trained even on CPUs. This network can be used to demonstrate the framework working. Additionally this acts a starting point for example/tutorial for getting started to know the Medical AI library.

call(inputSize, OutputSize, convLayers=None)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.tinyMedNet_v2()¶

Bases: medicalai.chief.networks.NetworkInit

tinyMedNet_v2 allows users to configure the number of Conv/CNN layers. tinyMedNet_v2 is a classification network that consumes very less resources and can be trained even on CPUs. This network can be used to demonstrate the framework working. Additionally this acts a starting point for example/tutorial for getting started to know the Medical AI library.

call(inputSize, OutputSize, convLayers=2)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

class medicalai.chief.networks.tinyMedNet_v3()¶

Bases: medicalai.chief.networks.NetworkInit

tinyMedNet_v3 has 3 FC layers with Dropout and Configurable number of Conv/CNN Layers.

call(inputSize, OutputSize, convLayers=2)¶

Sample should return model initialized with input and output Sizes.

Parameters
- inputSize (tuple** or **int.) – Integer or tuple specifying the input of network.
- OutputSize (tuple** or **int.) – Integer or tuple specifying the output classes of network.
Returns

Initialized Model.
Return type

numpy.array.

medicalai.chief.prettyloss module¶

class medicalai.chief.prettyloss.prettyLoss(show_percentage=False)¶

Bases: object