--- a
+++ b/utils.py
@@ -0,0 +1,174 @@
+import os
+import cv2
+import math
+import imutils
+
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+import config
+
+
+def resize(image, 
+           x=config.IMAGE_PXL_SIZE_X,
+           y=config.IMAGE_PXL_SIZE_Y):
+    if not len(image):
+        return np.array([])
+
+    return np.stack([cv2.resize(scan, (x, y)) 
+                     for scan in image])
+
+
+def normalize(image):
+    image = ((image - config.MIN_BOUND) / 
+             (config.MAX_BOUND - config.MIN_BOUND))
+    image[image>1] = 1.
+    image[image<0] = 0.
+    return image
+
+
+def chunks(l, n):
+    """Yield successive n-sized chunks from l."""
+    for i in range(0, len(l), n):
+        yield l[i:i + n]
+
+
+def mean(l):
+    if len(l):
+        return sum(l) / len(l)
+    return np.full(l.shape, config.OUT_SCAN, l.dtype)
+
+
+def get_mean_chunk_slices(slices):
+    if len(slices) < config.SLICES:
+        print("New slices are less then required after getting mean images, adding padding.")
+        return trim_and_pad(np.array(slices), config.SLICES) 
+
+    new_slices = []
+    for slice_chunk in np.array_split(slices, config.SLICES):
+        slice_chunk = list(map(mean, zip(*slice_chunk)))
+        new_slices.append(slice_chunk)
+
+    return np.stack(new_slices)
+
+
+def read_csv(input_file):
+    return pd.read_csv(input_file, index_col=0)
+
+
+def read_csv_column(input_file, columns=[0]):
+    return pd.read_csv(input_file, usecols=columns).values.flatten()
+
+
+def store_to_csv(patients, labels, csv_file_path):
+    index = pd.Index(data=patients, name=config.ID_COLUMN_NAME)
+    df = pd.DataFrame(data={config.COLUMN_NAME: labels}, 
+                      columns=[config.COLUMN_NAME],
+                      index=index)
+    df.to_csv(csv_file_path)
+
+
+def trim_and_pad(patient_img, slice_count, normalize_pad=True):
+    slices, size_x, size_y = patient_img.shape
+
+    if slices == slice_count:
+        return patient_img
+
+    if slices > slice_count:
+        return patient_img[:slice_count]
+
+    padding = np.full((slice_count-slices, size_x, size_y), 
+        config.OUT_SCAN, patient_img.dtype)
+    if normalize_pad:
+        padding = normalize(padding)
+
+    return np.vstack([patient_img, padding])
+
+
+def trim_pad_slices(scans, pad_with_existing=True,
+                    padding_value=config.BACKGROUND):
+    slices, x, y = scans.shape
+
+    if slices == config.SLICES:
+        return scans
+
+    if slices < config.SLICES:
+        pad = config.SLICES - slices
+        new_scans = []
+        if pad > slices:
+            # Double the size, scans are already ordered by slice location
+            for scan in scans:
+                new_scans.append(scan) 
+                new_scans.append(scan)
+
+            del scans
+            scans = new_scans
+            pad = config.SLICES - len(scans)
+    
+        if pad_with_existing:
+            padding = []
+            for slice_chunk in np.array_split(scans, pad):
+                padding.extend(slice_chunk)
+                padding.append(slice_chunk[-1])
+
+            #contains originals also, doubles the last slice in the chunk
+            return np.stack(padding)
+        else:
+            padding = np.full((pad, x, y), padding_value, scans.dtype)
+            return np.vstack([scans, padding])
+
+    trim = slices - config.SLICES
+    trimmed = []
+    for slice_chunk in np.array_split(scans, trim):
+        trimmed.append(slice_chunk[1:])
+
+    return np.vstack(trimmed)
+
+
+def count_background_rows(image, background=config.BACKGROUND):
+    return np.sum(np.all(image == background, axis=1))
+
+
+def remove_background_rows(image, background=config.BACKGROUND):
+    return image[40:image.shape[0]-40, 20:image.shape[1]-20]
+
+
+def rotate_scans(scans, angle=10):
+    return np.stack([imutils.rotate(scan, angle) for scan in scans])
+
+
+def remove_background_rows_3d(scans, background=config.BACKGROUND):
+    transformed = []
+    for scan in scans:
+        removed = remove_background_rows(scan, background)
+        tr_scan = cv2.resize(removed, 
+            (config.IMAGE_PXL_SIZE_X, config.IMAGE_PXL_SIZE_Y))
+        transformed.append(tr_scan)
+
+    return np.stack(transformed)
+
+
+def store_patient_image(image_dir, image, patient_id):
+    """
+    Serializes the patient image.
+
+    Image is a 3D numpy array - array from patient slices.
+    If not existing image_dir is created.
+    """
+    if not os.path.exists(image_dir):
+        os.makedirs(image_dir)
+
+    np.savez_compressed(os.path.join(image_dir, patient_id), image)
+
+
+def load_patient_image(image_dir, patient_id):
+    """
+    Load the serialized patient image.
+
+    Image is a 3D array - array of patient slices, metadata,
+    contained in the dicom format, is removed.
+    """
+    if '.npz' not in patient_id:
+        patient_id += '.npz'
+    with np.load(os.path.join(image_dir, patient_id)) as data:
+        return data['arr_0']
\ No newline at end of file