Applying MONAI transforms to sequences of images uniformly

Question

I am working with a medical dataset in which each datapoint is a sequence of 10 images. I want to apply MONAI transforms (Rand2DElastic, RandRotate, RandZoom, RandGaussianNoise) to these sequences for augmentation. These transforms should be applied randomly to each sequence, however each image in a given sequence should have the exact same transforms for consistency purposes. Is there already existing functionality to do so? If there is not, is there a good way of doing it? Data is stored as an np array of shape (n, 10, 3, 128, 128) Thanks!

images = [Image.open(os.path.join(self.root_dir, image_path)) for image_path in sequence_path]
images = [np.transpose(np.array(image), (2,0,1)) for image in images] 

        if self.augment:
            augment_transforms = Compose([
                Rand2DElastic(prob=0.6, spacing=(30, 30), magnitude_range=(0.1, 0.3)),
                RandRotate(range_x=np.pi / 60, prob=0.2, keep_size=True),
                RandZoom(min_zoom=0.8, max_zoom=1.5, prob=0.6),
                RandGaussianNoise(prob=0.5, mean=0, std=0.01)
            ])
            images = augment_transforms({"image": np.array(images)})["image"]
            print(images)

This is what I tried so far, but didn't give results I needed

Answer 1

i'm also working with MONAI right now. Maybe you can use the dictionary-transforms. I use them to do the same transformation on an image and the corresponding segmentation mask Tutorial for 3D seg. In your case, you should input your 10 Image series. Maybe another option is to concat your 10 objects to one object with one more dimension.