How to structure torchio.Subject for regression?

[sarthakpati]

Hi,

I am trying to use TorchIO for regression and I was wondering if there is any way to add the "value to predict" per subject and get it during the training cycle?

Here is what I have (for segmentation):

for patient in range(num_row):
    # We need this dict for storing the meta data for each subject such as different image modalities, labels, any other data
    subject_dict = {}
    # iterating through the channels/modalities/timepoints of the subject
    for channel in channelHeaders:
        # assigning the dict key to the channel
        subject_dict[str(channel)] = Image(str(dataframe[channel][patient]), type=torchio.INTENSITY)
    if labelHeader is not None:
        subject_dict['label'] = Image(str(dataframe[labelHeader][patient]), type=torchio.LABEL)
        if not train:
            subject_dict['path_to_metadata'] = str(dataframe[labelHeader][patient])
    else:
        subject_dict['label'] = "NA"
        if not train:
            subject_dict['path_to_metadata'] = str(dataframe[channel][patient])
    # Initializing the subject object using the dict
    subject = Subject(subject_dict) 
    # Appending this subject to the list of subjects
    subjects_list.append(subject)

Cheers,
Sarthak

[romainVala]

Hi
Have a look to Subject class

torchio/torchio/data/subject.py

Line 8 in 08af12b

class Subject(dict):

it gives an example where subject age in added at the init call
so you just need to add a key to subject_dict, ... is that what you are looking for ?

[fepegar]

@sarthakpati have you tried what @romainVala has suggested? Here's some code that might help:

In [10]: import torchio as tio

In [11]: subjects = [tio.Subject(t1=tio.ScalarImage('t1.nii.gz'), age=30+i) for i in range(5)]

In [12]: dataset = tio.SubjectsDataset(subjects)

In [13]: loader = DataLoader(dataset, batch_size=2, shuffle=True)

In [14]: for batch in loader:
    ...:     print(batch['age'])
    ...:
tensor([31, 33])
tensor([30, 32])
tensor([34])

Where the subject_dict syntax could have also been used.

[sarthakpati]

Hey, sorry I forgot to reply. Yes, I was able to get it working using @romainVala's suggestion.