diff --git a/README.md b/README.md
index d5a0f73..b1418a7 100644
--- a/README.md
+++ b/README.md
@@ -27,7 +27,8 @@ The software is currently composed of four separate loosely-coupled services. Sp
 The platform is designed so that it is possible to use `KNEEL` and `DeepKnee` separately. Both microservices expect
 a `JSON` with `{dicom: <I64>}`, where `<I64>` is the dicom file encoded in `base64`. If you make a request to either of the services,
 it needs to be done to `/kneel/predict/bilateral` or `/deepknee/predict/bilateral` for `KNEEL` and `DeepKnee`, respectively.
-A Jupyter Notebook `request_example.ipynb` demonstrates an example of such request.
+
+An example script that uses the platform can be found in the file `analyze_folder.py`.
 
 ## License
 This code is freely available only for research purposes. Commercial use is not allowed by any means.
diff --git a/analyze_folder.py b/analyze_folder.py
index 8a08579..c9be03d 100644
--- a/analyze_folder.py
+++ b/analyze_folder.py
@@ -17,17 +17,25 @@ def base64tonumpy(buffer):
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    parser.add_argument('--deepknee_host', default='http://127.0.0.1')
-    parser.add_argument('--deepknee_port', type=int, default=5001)
-    parser.add_argument('--img_dir', default='')
-    parser.add_argument('--save_results', default='/tmp/deepknee')
+    parser.add_argument('--deepknee_host', default='http://127.0.0.1', help='Host on which deepknee is running.')
+    parser.add_argument('--deepknee_port', type=int, default=5001, help='Port of deepknee.')
+    parser.add_argument('--img_dir', default='', help='Directory with images.')
+    parser.add_argument('--patient_level', type=bool, default=False,
+                        help='Whether there image lies in a patient directory. '
+                             'In this case, the script will know that the img_dir '
+                             'contains folders that contain images.')
+    parser.add_argument('--save_results', default='/tmp/deepknee', help='Folder where to save the results.')
     args = parser.parse_args()
 
     url = f'{args.deepknee_host}:{args.deepknee_port}/deepknee/predict/bilateral'
     os.makedirs(args.save_results, exist_ok=True)
     output_csv = []
-    flist = glob.glob(os.path.join(args.img_dir, '*'))
-    for img_path in tqdm(flist, total=len(flist)):
+    if not args.patient_level:
+        flist = glob.glob(os.path.join(args.img_dir, '*'))
+    else:
+        flist = glob.glob(os.path.join(args.img_dir, '*', '*'))
+
+    for idx, img_path in tqdm(enumerate(flist), total=len(flist)):
         # Encoding the DICOM as base64 and sending the request to the server
         with open(img_path, 'rb') as f:
             data_base64 = base64.b64encode(f.read()).decode('ascii')
@@ -39,9 +47,12 @@ def base64tonumpy(buffer):
             # You can also access the localized image, heatmaps and the probability maps
             result[knee] = {'img': base64tonumpy(res[knee]['img']),
                             'hm': base64tonumpy(res[knee]['hm']),
-                            'preds_bar': base64tonumpy(res[knee]['preds_bar']),
+                            'probs_bar': base64tonumpy(res[knee]['preds_bar']),
                             'kl': res[knee]['kl']}
             output_csv.append({'File': img_path, 'Side': knee, 'KL': result[knee]['kl']})
+            cv2.imwrite(os.path.join(args.save_results, f'{idx}_{knee}_img.png'), result[knee]['img'])
+            cv2.imwrite(os.path.join(args.save_results, f'{idx}_{knee}_hm.png'), result[knee]['hm'])
+            cv2.imwrite(os.path.join(args.save_results, f'{idx}_{knee}_probs.png'), result[knee]['probs_bar'])
 
     df = pd.DataFrame(data=output_csv)
     df.to_csv(os.path.join(args.save_results, 'deepknee.csv'), index=None)