Auto3DSeg Output came out as empty?

[jonathanquantumalpha]

Hi I have the following code in Colab for Auto3DSeg, but I am not sure why the result comes out as empty (background/foreground for both groundtruth/prediction)? I imported the Lung dataset from Medical Decathlon and imported part of it to my Google Drive (I checked the folder locations and they are correct).

import os
import json
import nibabel as nib
import numpy as np
import matplotlib.pyplot as plt
import torch

from monai.apps.auto3dseg import AutoRunner
from monai.config import print_config
from monai.data import create_test_image_3d

print_config()

from monai.utils import first, set_determinism
from monai.losses import DiceLoss, DiceCELoss
from monai.transforms import (
    AsDiscrete,
    AsDiscreted,
    EnsureChannelFirstd,
    Compose,
    CropForegroundd,
    LoadImaged,
    Orientationd,
    RandCropByPosNegLabeld,
    SaveImaged,
    ScaleIntensityRanged,
    Spacingd,
    Invertd,
    AddChanneld,
    Resized,
    ScaleIntensityd,
    Resized,
    RandAffined,
    ToTensord
)

from monai.networks.nets import UNet
from monai.networks.layers import Norm
from monai.data import CacheDataset, DataLoader, Dataset, decollate_batch

import torch
import matplotlib.pyplot as plt

import os
from glob import glob
import numpy as np

from monai.inferers import sliding_window_inference
from monai.metrics import DiceMetric
import tempfile
import random

sim_datalist = {
    "testing": [
        {"image": "lung_074.nii.gz", "label": "lung_074.nii.gz"},
        {"image": "lung_075.nii.gz", "label": "lung_075.nii.gz"},
        {"image": "lung_078.nii.gz", "label": "lung_078.nii.gz"},
        {"image": "lung_079.nii.gz", "label": "lung_079.nii.gz"},
        {"image": "lung_080.nii.gz", "label": "lung_080.nii.gz"},
        {"image": "lung_081.nii.gz", "label": "lung_081.nii.gz"},
        {"image": "lung_083.nii.gz", "label": "lung_083.nii.gz"},
        {"image": "lung_084.nii.gz", "label": "lung_084.nii.gz"},
        {"image": "lung_086.nii.gz", "label": "lung_086.nii.gz"},
        {"image": "lung_092.nii.gz", "label": "lung_092.nii.gz"},
        {"image": "lung_093.nii.gz", "label": "lung_093.nii.gz"},
        {"image": "lung_095.nii.gz", "label": "lung_095.nii.gz"},
        {"image": "lung_096.nii.gz", "label": "lung_096.nii.gz"},
    ],
    "training": [
        {"fold": 0, "image": "lung_001.nii.gz", "label": "lung_001.nii.gz"},
        {"fold": 0, "image": "lung_003.nii.gz", "label": "lung_003.nii.gz"},
        {"fold": 0, "image": "lung_004.nii.gz", "label": "lung_004.nii.gz"},
        {"fold": 1, "image": "lung_005.nii.gz", "label": "lung_005.nii.gz"},
        {"fold": 1, "image": "lung_006.nii.gz", "label": "lung_006.nii.gz"},
        {"fold": 1, "image": "lung_009.nii.gz", "label": "lung_009.nii.gz"},
        {"fold": 2, "image": "lung_010.nii.gz", "label": "lung_010.nii.gz"},
        {"fold": 2, "image": "lung_014.nii.gz", "label": "lung_014.nii.gz"},
        {"fold": 2, "image": "lung_015.nii.gz", "label": "lung_015.nii.gz"},
        {"fold": 3, "image": "lung_016.nii.gz", "label": "lung_016.nii.gz"},
        {"fold": 3, "image": "lung_018.nii.gz", "label": "lung_018.nii.gz"},
        {"fold": 3, "image": "lung_020.nii.gz", "label": "lung_020.nii.gz"},
        {"fold": 4, "image": "lung_022.nii.gz", "label": "lung_022.nii.gz"},
        {"fold": 4, "image": "lung_023.nii.gz", "label": "lung_023.nii.gz"},
        {"fold": 4, "image": "lung_025.nii.gz", "label": "lung_025.nii.gz"},
        {"fold": 5, "image": "lung_026.nii.gz", "label": "lung_026.nii.gz"},
        {"fold": 5, "image": "lung_027.nii.gz", "label": "lung_027.nii.gz"},
        {"fold": 5, "image": "lung_028.nii.gz", "label": "lung_028.nii.gz"},
        {"fold": 6, "image": "lung_029.nii.gz", "label": "lung_029.nii.gz"},
        {"fold": 6, "image": "lung_031.nii.gz", "label": "lung_031.nii.gz"},
        {"fold": 6, "image": "lung_033.nii.gz", "label": "lung_033.nii.gz"},
        {"fold": 7, "image": "lung_034.nii.gz", "label": "lung_034.nii.gz"},
        {"fold": 7, "image": "lung_036.nii.gz", "label": "lung_036.nii.gz"},
        {"fold": 7, "image": "lung_037.nii.gz", "label": "lung_037.nii.gz"},
        {"fold": 8, "image": "lung_038.nii.gz", "label": "lung_038.nii.gz"},
        {"fold": 8, "image": "lung_041.nii.gz", "label": "lung_041.nii.gz"},
        {"fold": 8, "image": "lung_042.nii.gz", "label": "lung_042.nii.gz"},
        {"fold": 9, "image": "lung_043.nii.gz", "label": "lung_043.nii.gz"},
        {"fold": 9, "image": "lung_044.nii.gz", "label": "lung_044.nii.gz"},
        {"fold": 9, "image": "lung_045.nii.gz", "label": "lung_045.nii.gz"},
        {"fold": 10, "image": "lung_046.nii.gz", "label": "lung_046.nii.gz"},
        {"fold": 10, "image": "lung_047.nii.gz", "label": "lung_047.nii.gz"},
        {"fold": 10, "image": "lung_049.nii.gz", "label": "lung_049.nii.gz"},
        {"fold": 11, "image": "lung_051.nii.gz", "label": "lung_051.nii.gz"},
        {"fold": 11, "image": "lung_053.nii.gz", "label": "lung_053.nii.gz"},
        {"fold": 11, "image": "lung_054.nii.gz", "label": "lung_054.nii.gz"},
        {"fold": 12, "image": "lung_055.nii.gz", "label": "lung_055.nii.gz"},
        {"fold": 12, "image": "lung_057.nii.gz", "label": "lung_057.nii.gz"},
        {"fold": 12, "image": "lung_058.nii.gz", "label": "lung_058.nii.gz"},
        {"fold": 13, "image": "lung_059.nii.gz", "label": "lung_059.nii.gz"},
        {"fold": 13, "image": "lung_061.nii.gz", "label": "lung_061.nii.gz"},
        {"fold": 13, "image": "lung_062.nii.gz", "label": "lung_062.nii.gz"},
        {"fold": 14, "image": "lung_064.nii.gz", "label": "lung_064.nii.gz"},
        {"fold": 14, "image": "lung_065.nii.gz", "label": "lung_065.nii.gz"},
        {"fold": 14, "image": "lung_066.nii.gz", "label": "lung_066.nii.gz"},
        {"fold": 15, "image": "lung_069.nii.gz", "label": "lung_069.nii.gz"},
        {"fold": 15, "image": "lung_070.nii.gz", "label": "lung_070.nii.gz"},
        {"fold": 15, "image": "lung_071.nii.gz", "label": "lung_071.nii.gz"},
        {"fold": 16, "image": "lung_073.nii.gz", "label": "lung_073.nii.gz"},
    ],
}

sim_dim = (64, 64, 64)

if torch.cuda.device_count() > 4:
    os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
    os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3"

from google.colab import drive
drive.mount('/content/drive')

directory = os.environ.get("MONAI_DATA_DIRECTORY")
root_dir = tempfile.mkdtemp() if directory is None else directory
print(root_dir)

data_dir = '/content/drive/MyDrive/Task06_Lung'

train_images = sorted(glob(os.path.join(data_dir, "TrainData", "*.nii.gz")))
train_labels = sorted(glob(os.path.join(data_dir, "TrainLabels", "*.nii.gz")))

val_images = sorted(glob(os.path.join(data_dir, "ValData", "*.nii.gz")))
val_labels = sorted(glob(os.path.join(data_dir, "ValLabels", "*.nii.gz")))

train_files = [{"fold": random.randint(0,5), "image": image_name, "label": label_name} for image_name, label_name in zip(train_images, train_labels)]
val_files = [{"image": image_name, "label": label_name} for image_name, label_name in zip(val_images, val_labels)]

lung_datalist={"testing": val_files, "training": train_files}
print(lung_datalist)

print(train_files )

img_nib = nib.load(lung_datalist['training'][2]['image'])
lbl_nib = nib.load(lung_datalist['training'][2]['label'])
# img_nib = nib.load(lung_datalist['training'][1]['image'])
# lbl_nib = nib.load(lung_datalist['training'][1]['label'])
img = np.array(img_nib.dataobj)
lbl = np.array(lbl_nib.dataobj)

plt.subplot(1, 2, 1)
plt.imshow(img[:,:,132])
plt.title('image')
cbar = plt.colorbar(shrink=0.55)
plt.subplot(1, 2, 2)
plt.imshow(lbl[:,:,132])
plt.title('label')
cbar = plt.colorbar(shrink=0.55)

work_dir='/content/drive/MyDrive/Task06_Lung/'
datalist_file = os.path.join(work_dir, "lung_datalist.json")

# Create the results directory if it doesn’t exist
results_dir = os.path.join(work_dir, 'results')
os.makedirs(results_dir, exist_ok=True)
# Create the path for the JSON file
datalist_file = os.path.join(results_dir, 'lung_datalist.json')
# Write the ‘lung_datalist’ to the JSON file
with open(datalist_file, 'w') as f:
    json.dump(lung_datalist, f)

runner = AutoRunner(
    work_dir=work_dir,
    # work_dir=results_dir,
    input={
        "modality": "CT",
        "datalist": datalist_file,
        "dataroot": '',
    },
    algos='segresnet',
)

max_epochs = 300

# safeguard to ensure max_epochs is greater or equal to 2
max_epochs = max(max_epochs, 2)

train_param = {
    "CUDA_VISIBLE_DEVICES": [0],  # use only 1 gpu
    "num_iterations": 4 * max_epochs,
    "num_iterations_per_validation": 2 * max_epochs,
    "num_images_per_batch": 2,
    "num_epochs": max_epochs,
    "num_warmup_iterations": 2 * max_epochs,
}
runner.set_training_params(train_param)
runner.set_num_fold(num_fold=1)

runner.run()

image_name = sim_datalist['testing'][0]['image'].split(".")[0]
prediction_nib = nib.load(os.path.join(work_dir, 'ensemble_output', image_name+"_ensemble"+".nii.gz"))
pred = np.array(prediction_nib.dataobj)

img_slice32 = lbl[32] == 0
label_slice32 = lbl[32] == 1
background_slice32 = pred[32, :, :, 0] if pred.ndim == 4 else pred[32] == 0
foreground_slice32 = pred[32, :, :, 1] if pred.ndim == 4 else pred[32] == 1

plt.subplot(2, 2, 1)
plt.imshow(img_slice32)
plt.title('background groundtruth')
cbar = plt.colorbar(shrink=0.8)
plt.subplot(2, 2, 2)
plt.imshow(label_slice32)
plt.title('foreground_groundtruth')
cbar = plt.colorbar(shrink=0.8)
plt.subplot(2, 2, 3)
plt.imshow(background_slice32)
plt.title('background prediction')
cbar = plt.colorbar(shrink=0.8)
plt.subplot(2, 2, 4)
plt.imshow(foreground_slice32)
plt.title('foreground prediction')
cbar = plt.colorbar(shrink=0.8)
# set the spacing between subplots
plt.subplots_adjust(left=0.1,
                    bottom=0.1,
                    right=0.9,
                    top=0.9,
                    wspace=0.4,
                    hspace=0.4)
plt.show()

# Loop through all items in sim_datalist['testing']
for item in sim_datalist['testing']:
    image_name = item['image'].split(".")[0]
    prediction_nib = nib.load(os.path.join(work_dir, 'ensemble_output', image_name + "_ensemble" + ".nii.gz"))
    pred = np.array(prediction_nib.dataobj)

    for i in range(50):
      img_slice32 = lbl[i] == 0
      label_slice32 = lbl[i] == 1
      background_slice32 = pred[i, :, :, 0] if pred.ndim == 4 else pred[i] == 0
      foreground_slice32 = pred[i, :, :, 1] if pred.ndim == 4 else pred[i] == 1

      plt.subplot(2, 2, 1)
      plt.imshow(img_slice32)
      plt.title('background groundtruth')
      cbar = plt.colorbar(shrink=0.8)
      plt.subplot(2, 2, 2)
      plt.imshow(label_slice32)
      plt.title('foreground_groundtruth')
      cbar = plt.colorbar(shrink=0.8)
      plt.subplot(2, 2, 3)
      plt.imshow(background_slice32)
      plt.title('background prediction')
      cbar = plt.colorbar(shrink=0.8)
      plt.subplot(2, 2, 4)
      plt.imshow(foreground_slice32)
      plt.title('foreground prediction')
      cbar = plt.colorbar(shrink=0.8)
      # set the spacing between subplots
      plt.subplots_adjust(left=0.1,
                          bottom=0.1,
                          right=0.9,
                          top=0.9,
                          wspace=0.4,
                          hspace=0.4)
      plt.show()

[KumoLiu]

Hi @jonathanquantumalpha, could you please refer to these tutorials as the first step?
https://github.com/Project-MONAI/tutorials/tree/main/auto3dseg
https://github.com/Project-MONAI/tutorials/blob/main/auto3dseg/notebooks/auto3dseg_hello_world.ipynb
Thanks!