Results & Discussion: Age Prediction Model for Individuals with Schizophrenia
by Lilo Ryan
The results are unfortunately not as exciting as were initially expected as a result of poor coding.
The image in Figure 1 helps to visualize the kind of data that is being inputted to the model. The white areas are the fatty tissues.
The MRI data had to be cropped, but not severely edited— the data came in a skull-stripped format for T1W format. There is a major difference between the T1W and T2W formats— T1 images highlight fat with their radiofrequency pulse sequences (Lloyd-Jones 2017). T2W images highlight both fat and water in the representation of images. T2W images indicate very clearly different kinds of lesions in the brain, which makes them incredibly valuable in the medical field. For the pyment module and the testing performed in this experiment, we want to use the T1 images so as to not have any blurring of the ideal signals in the brain. In addition, the preprocessing tools that we use to get the data up to expectation are only available for the T1 images. The datasets actually have a reasonable number of samples for the test set– it was shocking to find such a robust dataset for human MRI data.
| Control | Schizophrenia | |
| Test size | 19 | 31 |
So, how are these MRI signals processed to be used in the model? The format that is most commonly used is called a “Nifti” file— this filetype is primarily used in the MRI processing community. The files were prepared using the library that processes Nifti images— nibabel.
In terms of the results, they were unfortunately difficult to attain. Unfortunately, the only outputs produced were 5 and 65. There is likely a deeper issue in my code that means that it cannot be actually similar enough to the training set to accurately predict anything.
In the process of working on generating code from this test dataset, there were many different stumbling blocks. In the actual downloading of the files, there were frequent issues with getting the different files directly onto the HPC. I learned the value of not downloading files to an intermediate device and rather to go directly to the HPC.
I failed to realize the necessity of the preprocessing until extremely late into the processing/data. I got the code from the pyment model working (mostly) and simplified the imports significantly. I used pathlib to get the files and loop through. The preprocessing steps involve cropping, fitting to a template image, reorienting the image, transforming, and finally applying the brainmask.
Next steps on the project would be to actually get the images read in and correctly processed. Verifying the images as compared to the training set (which was attempted) would be a good way to see if the inputs are actually correct.
References
- “About the Mni Space(s).” About the MNI Space(s), Lead Labs, www.lead-dbs.org/about-the-mni-spaces/. Accessed 7 May 2025.
- Dörfel, Ruben P., et al. “Prediction of brain age using structural magnetic resonance imaging: A comparison of accuracy and test–retest reliability of publicly available software packages.” Human Brain Mapping, vol. 44, no. 17, 16 Oct. 2023, pp. 6139–6148, https://doi.org/10.1002/hbm.26502.
- Keane, Brian P, et. al (2024). The Neural Basis of Visual Shape Completion in Schizophrenia and Bipolar Disorder. OpenNeuro. [Dataset] doi: doi:10.18112/openneuro.ds005073.v1.0.0
- Keane, Brian P., et al. “Dorsal attention network activity during perceptual organization is distinct in schizophrenia and predictive of cognitive disorganization.” European Journal of Neuroscience, vol. 57, no. 3, 4 Jan. 2023, pp. 458–478, https://doi.org/10.1111/ejn.15889.
- Lloyd-Jones , Graham. “MRI Interpretation.” Radiology Masterclass, Salisbury NHS Foundation Trust UK, Sept. 2017, www.radiologymasterclass.co.uk/tutorials/mri/t1_and_t2_images.
- Snoek, Lukas. “Working with MRI Data in Python.” Working with MRI Data in Python – NI-Edu, 2021, lukas-snoek.com/NI-edu/fMRI-introduction/week_1/python_for_mri.html.
- “FIRST/UserGuide”, Harvard, 2014, ftp.nmr.mgh.harvard.edu/pub/dist/freesurfer/tutorial_packages/centos6/fsl_507/doc/wiki/FIRST(f)UserGuide.html.