dicom.py

if file.filename.endswith('.jpg') or file.filename.endswith('.png'):
    is_dicom = False
    img = cv2.imdecode(np.fromstring(file.read(), np.uint8), 1)
else:
    is_dicom = True
    df = dicom.dcmread(file)  # .pixel_array

    slope = float(df.RescaleSlope)
    intercept = float(df.RescaleIntercept)
    df_data = intercept + df.pixel_array * slope / 20

    img = df_data.astype(np.float32)
    img = 255 - img
    img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)

model = load_model(MODEL_NAME)

img = cv2.resize(img, (IMAGE_HEIGHT, IMAGE_WIDTH))
img = img.astype("float") / 255.0
img = img_to_array(img)
img = np.expand_dims(img, axis=0)

classes = model.predict(img)