def directory_iterator(root):
"""
Iterator through a directory that contains:
1) individual subdirectories that contain SRH strips in alterating order
2) Bijection of specimens to directories
"""
pred_dict = defaultdict(list)
for dirpath, dirname, files in os.walk(root):
if "NIO" in dirpath:
try:
mosaic = import_srh_dicom(dirpath)
patches = patch_generator(mosaic)
normalized_dist = prediction(feedforward(patches, model))
pred_dict["specimen"].append(dirpath.split("/")[-1]) # select only the filename, remove root
# append the predictions to each class key in the prediction dictionary
for i in range(len(CLASS_NAMES)):
pred_dict[CLASS_NAMES[i]].append(normalized_dist[i])
except:
continue
return pred_dict
def directory_iterator(root):
mosaic_dirs = os.listdir(root)
mosaic_dirs = sorted(mosaic_dirs)
print(mosaic_dirs)
counter = 0
mosaic_dict = {}
for mosaic_dir in mosaic_dirs:
counter += 1
mosaic = import_srh_dicom(os.path.join(root, mosaic_dir))
patches = patch_generator(mosaic)
inference_mah = feedforward(patches, model=output_model)
score_dict = mahalanobis_score_ensemble(maha_class, inference_mah)
mosaic_dict[mosaic_dir] = score_dict
print(counter)
return mosaic_dict
plt.imshow(heatmap[:,:,class_index])
plt.show()
def save_heatmap_image(file_name, heatmap, class_index, cmap='Greys_r'):
"""
Saves the greyscale heatmap image for specific class
"""
plt.imsave(file_name, heatmap[:,:,class_index], cmap='Greys_r', vmin=0,vmax=1)
if __name__ == '__main__':
# import model and specify image directory
model = load_model("/home/todd/Desktop/RecPseudo_project/patches/cv_round2/recurmodel_kthfold_2.hdf5")
image_dir = "/home/todd/Desktop/Recurrence_mosaics_figures/NIODS10070_2"
# import image
mosaic = import_srh_dicom(image_dir)
image_size = mosaic.shape[0]
# index map to allow for patch-tile mapping
heatmap_indices = indices_map(mosaic, step_size = STEP_SIZE)
# predict on each patch
patch_dict = patch_dictionary(mosaic, model, heatmap_indices, image_preprocced=False, step_size = STEP_SIZE)
# generate heatmap predictions
heatmap = srh_heatmap(patch_dict, image_size, step_size = STEP_SIZE)
plt.imshow(heatmap[:,:,1])
plt.show()
output_model = mahalanobis_model(model_path)
val_batch, val_steps = validation_batch_steps(training_dir)
train_generator = ImageDataGenerator(
horizontal_flip=True,
vertical_flip=True,
preprocessing_function=cnn_preprocessing,
data_format="channels_last").flow_from_directory(
directory=training_dir,
target_size=(img_rows, img_cols),
color_mode='rgb',
classes=None,
class_mode='categorical',
batch_size=val_batch,
shuffle=False)
mahalanobis_output = output_model.predict_generator(train_generator,
steps=val_steps,
verbose=1)
maha_class = Class_conditional_dicts(layer_list=layer_outputs)
maha_class.populate_mu_sigmas(mahalanobis_output, train_generator)
mosaic_dict = directory_iterator("")
export_mahalanobis_scores(mosaic_dict, layer_outputs, metric="mean")
mosaic = import_srh_dicom("")
patches = patch_generator(mosaic)
inference_mah = feedforward(patches, model=output_model)
mosaic_dict = mahalanobis_score_ensemble(maha_class, inference_mah)
args = parser.parse_args()
CLASS_NAMES = [
'ependymoma', 'greymatter', 'glioblastoma', 'lowgradeglioma', 'lymphoma',
'medulloblastoma', 'meningioma', 'metastasis', 'nondiagnostic',
'pilocyticastrocytoma', 'pituitaryadenoma', 'pseudoprogression',
'schwannoma', 'whitematter'
]
if __name__ == "__main__":
# load a trained SRH model
deepSRHmodel = load_model(args.model)
# import SRH mosaic
specimen = import_srh_dicom(args.strip_directory)
print("SRH mosaic size is: " + str(specimen.shape))
# generate preprocessed image patches
patches = patch_generator(specimen, step_size=100, old_preprocess=True)
del specimen
# predict on patches
specimen_prediction = prediction(
feedforward(patch_dict=patches, model=deepSRHmodel))
# print diagnosis
print("SRH specimen diagnosis is: " +
CLASS_NAMES[np.argmax(specimen_prediction)] + " (probability = " +
str(np.max(specimen_prediction)) + ")")
Function to return a list of directories that contain raw SRH strips
"""
dir_list = sorted(os.listdir(parent_dir))
directory_list = []
for directory in dir_list:
directory_list.append(os.path.join(parent_dir, directory))
return directory_list
if __name__ == "__main__":
# root directory with directories of STRIPS from single mosaic
src_dir = "/home/todd/Desktop/test_strips"
# destination for ALL PATCHES
dest_dir = "/home/todd/Desktop/test_patch_generator"
# ordered list of strip directories
ordered_dir_list = directory_list(src_dir)
for strip_directory in ordered_dir_list:
print(strip_directory)
# import mosaic
mosaic = import_srh_dicom(strip_directory, flatten=False)
# returns the patches to save and patches to pass to CNN
patches = patch_generator(mosaic, step_size=200)
# save patches
os.chdir(dest_dir)
patch_saver(patches, src_dir=strip_directory)