def route_api_query_image():
"""
api endpoint to query an image
this is the 'main' api endpoint for the application
reads two pieces of data from the request: "data" and "file"
the file is the binary image
the data is json payload of other request data,
which is just the lesion point clicked on by the use
Returns:
json response of:
the two probabilites from two models
the filtered and segmented image
the extracted patch
list of similar images
"""
if not auth_check(request.authorization):
return auth_fail()
request_data = json.loads(request.form["data"])
f = request.files["file"]
pil_img = Image.open(f)
img = cv2.cvtColor(np.array(pil_img), cv2.COLOR_RGB2BGR)
filtered_img = preprocess.preprocess(img)
filtered_img = g_data["segmenter"].segmenter(filtered_img)
translated_patch_coordinates = preprocess.translate_patch_coordinates(
filtered_img, request_data["point"])
patch = util.extract_patch(
filtered_img,
(translated_patch_coordinates["y"], translated_patch_coordinates["x"]),
preprocess.PATCH_SIZE)
patch = preprocess.preprocess(patch)
filtered_img = preprocess.preprocess(filtered_img)
prob1 = g_data["classifier1"].classify(filtered_img)
prob2 = g_data["classifier2"].classify(patch)
prob3 = g_data["classifier3"].classify(filtered_img)
if math.isnan(prob2):
prob2 = None
similarities = g_data["hash_similarity"].query_image(patch)
similarities = g_data["labels"].add_labels_to_similarity_list(similarities)
img_b64 = img_to_base64(filtered_img)
patch_b64 = img_to_base64(patch)
res = {
"filtered_img": img_b64.decode("utf-8"),
"patch": patch_b64.decode("utf-8"),
"probability1": prob1,
"probability2": prob2,
"probability3": prob3,
"similar_images": similarities,
}
return jsonify(res)
def main():
"""
this script automatically segments each images using the lung segmentation model
CLI Args:
1: the path to the lung segmentation model file
2: the path to the directory with all the images that will be segmented
3: the path to the output directory where all the segmented images will be written with the same filename
"""
model_file = sys.argv[1]
image_dir = sys.argv[2]
out_dir = sys.argv[3]
graph = tf.get_default_graph()
segmenter = Segmenter(model_file, graph)
filenames = os.listdir(image_dir)
for fn in filenames:
img = Image.open(opj(image_dir, fn))
#img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
img = preprocess.preprocess(np.array(img))
img = segmenter.segmenter(img)
img = Image.fromarray(np.uint8(img * 255))
img = img.convert("L")
img.save(opj(out_dir, fn))
def main():
"""
this script creates a results.csv in the output directory, which is used by the test-model-stats.py script for creating results graphs and statistics
this scripts is for testing the VGG16 based model
CLI Args:
1: the labels file for the test set of subjects
2: the directory of PNG image slices
3: the output directory for the results
4: the classifier model file
5: the lung segmentation model file
"""
test_labels = sys.argv[1]
image_dir = sys.argv[2]
out_dir = sys.argv[3]
model_file = sys.argv[4]
seg_model_file = sys.argv[5]
graph = tf.get_default_graph()
segmenter = model.Segmenter(seg_model_file, graph)
classifier = model.Classifier2(model_file, graph)
file_data = []
with open(test_labels) as f:
reader = csv.reader(f)
header = next(reader)
header.append("Prediction")
for line in reader:
fn = line[0]
img = Image.open(opj(image_dir, fn + ".png"))
img = preprocess.preprocess(np.array(img))
img = segmenter.segmenter(img)
img = preprocess.preprocess(img)
prediction = classifier.classify(img)
line.append(prediction)
file_data.append(line)
with open(opj(out_dir, "results.csv"), "w") as f:
writer = csv.writer(f)
writer.writerow(header)
for line in file_data:
writer.writerow(line)