I'm trying to create a Tensorflow Dataset that loads the DICOM images one batch at a time, but I'm running into issues. I started by loading (in the dataset pipeline) the images using tfio.image.decode_dicom_image(path), but every time an image is loaded it prints \"W: DcmMetaInfo: No Group Length available in Meta Information Header\", which makes training difficult.
\nI then tried pydicom.dcmread(path).pixel_array in the pipeline, but that runs into an error, as the dataset pipeline's graph turns the path into a tensor of dtype string (as opposed to just a string) that the dcmread function cannot accept.
\nHelp with either of these two issues would be appreciated.
", "messages": [ { "id": 1980332, "postDate": "2022-10-10T04:59:49.147Z", "content": "I also get the same warning message when loading these DCM files using tensorflow io. I couldn't figure a way to silence those messages. Here is a code snippet that I use to load DCM files
\ndef load_dicom(path):\n img=dicom.dcmread(path)\n img.PhotometricInterpretation = 'YBR_FULL'\n data = img.pixel_array \n data = data - np.min(data)\n if np.max(data) != 0:\n data = data / np.max(data)\n data = data.astype(np.float32)\n return data\n\ndef px_array(filename):\n return tf.convert_to_tensor(load_dicom(filename.numpy().decode(\"utf-8\") ))\n\ndef read_dcm_tf(filename):\n img = tf.py_function(px_array, inp=[filename], Tout=tf.float32) \n img = tf.stack([img,img,img], axis = 2)\n img = tf.image.resize(img, [IMAGE_SIZE, IMAGE_SIZE])\n return img\n\ndef process_image(filename, target=None):\n image = read_dcm_tf(filename)\n if (target is not None):\n return image, target\n else:\n return image\n\nds1 = (tf.data.Dataset.from_tensor_slices((your_list_of_dcm_files))\n .map(process_image, num_parallel_calls=tf.data.experimental.AUTOTUNE)\n .cache()\n .batch(16)\n .prefetch(tf.data.experimental.AUTOTUNE))\n\ndef display_batch(batch, size=2):\n imgs = batch\n for img_idx in range(size):\n plt.imshow(imgs[img_idx,...].numpy())\n plt.show()\n\n\nds = ds1.unbatch().batch(4)\nbatch = next(iter(ds1))\ndisplay_batch(batch, 2)\nI'm trying to create a Tensorflow Dataset that loads the DICOM images one batch at a time, but I'm running into issues. I started by loading (in the dataset pipeline) the images using tfio.image.decode_dicom_image(path), but every time an image is loaded it prints \"W: DcmMetaInfo: No Group Length available in Meta Information Header\", which makes training difficult.
\nI then tried pydicom.dcmread(path).pixel_array in the pipeline, but that runs into an error, as the dataset pipeline's graph turns the path into a tensor of dtype string (as opposed to just a string) that the dcmread function cannot accept.
\nHelp with either of these two issues would be appreciated.
", "rawMarkdown": "I'm trying to create a Tensorflow Dataset that loads the DICOM images one batch at a time, but I'm running into issues. I started by loading (in the dataset pipeline) the images using tfio.image.decode_dicom_image(path), but every time an image is loaded it prints \"W: DcmMetaInfo: No Group Length available in Meta Information Header\", which makes training difficult. \n\nI then tried pydicom.dcmread(path).pixel_array in the pipeline, but that runs into an error, as the dataset pipeline's graph turns the path into a tensor of dtype string (as opposed to just a string) that the dcmread function cannot accept.\n\nHelp with either of these two issues would be appreciated.", "votes": 1 } ], "comments": [ { "id": 1980332, "author_name": "Parag", "author_url": "", "post_date": "2022-10-10T04:59:49.147000", "content": "I also get the same warning message when loading these DCM files using tensorflow io. I couldn't figure a way to silence those messages. Here is a code snippet that I use to load DCM files
\ndef load_dicom(path):\n img=dicom.dcmread(path)\n img.PhotometricInterpretation = 'YBR_FULL'\n data = img.pixel_array \n data = data - np.min(data)\n if np.max(data) != 0:\n data = data / np.max(data)\n data = data.astype(np.float32)\n return data\n\ndef px_array(filename):\n return tf.convert_to_tensor(load_dicom(filename.numpy().decode(\"utf-8\") ))\n\ndef read_dcm_tf(filename):\n img = tf.py_function(px_array, inp=[filename], Tout=tf.float32) \n img = tf.stack([img,img,img], axis = 2)\n img = tf.image.resize(img, [IMAGE_SIZE, IMAGE_SIZE])\n return img\n\ndef process_image(filename, target=None):\n image = read_dcm_tf(filename)\n if (target is not None):\n return image, target\n else:\n return image\n\nds1 = (tf.data.Dataset.from_tensor_slices((your_list_of_dcm_files))\n .map(process_image, num_parallel_calls=tf.data.experimental.AUTOTUNE)\n .cache()\n .batch(16)\n .prefetch(tf.data.experimental.AUTOTUNE))\n\ndef display_batch(batch, size=2):\n imgs = batch\n for img_idx in range(size):\n plt.imshow(imgs[img_idx,...].numpy())\n plt.show()\n\n\nds = ds1.unbatch().batch(4)\nbatch = next(iter(ds1))\ndisplay_batch(batch, 2)\n