The effect of preprocessing filters on predictive performance in radiomics

In radiomics, the main goal is to extract quantitative features from medical data to train a predictive model using machine learning techniques. Contrary to statistics, radiomics is data-driven; thus, there is a certain tendency to use as many features as possible. To achieve this, preprocessing filters are often applied; for example, a Gaussian filter smooths the image and thus removes noise, and a wavelet filter decomposes the image into spectral frequencies that features can utilize.

Yet, this approach comes with a price. Many of these features will be irrelevant and redundant, complicating the training. It is well known that increasing the number of features comes hand in hand with a high risk of overfitting, especially if there are many more features than samples, which is often the case in radiomics. This effect is also called the curse of dimensionality. Therefore, adding many features could decrease predictive performance, contradicting the original intention.

Our study tested the effect of adding preprocessed features on predictive performance. On seven publicly available datasets, we compared the impact of eight different preprocessing filters on the area-under-the-curve (AUC). We found that using preprocessing filters can result in significant improvements of up to +0.08 in AUC compared to using only the original features. However, though it also decreased performance (up to -0.10) on some data sets, these drops were not statistically significant. Tuning the set of preprocessing filters always led to increased performance, although failing to reach statistical significance.

Based on these results, we recommend using all preprocessing filters; using too few features might hurt predictive performance. Ideally, the set of features should be tuned during training.

Key points

  • Modeling using all preprocessed features did not reduce the predictive performance compared to the original (not preprocessed) set.
  • Tuning image processing filters during cross-validation improves the performance slightly.
  • Using all preprocessed features seemed to be more helpful for datasets with larger sample size.
  • Pairwise Pearson correlations of preprocessed feature sets were high (r > 0.49).

Article: The effect of preprocessing filters on predictive performance in radiomics

Authors: Aydin Demircioğlu


  • Aydin Demircioğlu

    Institute of Diagnostic and Interventional Radiology and Neuroradiology, University Hospital Essen, Essen, Germany

Latest posts

Become A Member Today!

You will have access to a wide range of benefits that can help you advance your career and stay up-to-date with the latest developments in the field of radiology. These benefits include access to educational resources, networking opportunities with other professionals in the field, opportunities to participate in research projects and clinical trials, and access to the latest technologies and techniques. 

Check out our different membership options.

If you don’t find a fitting membership send us an email here.


for radiologists, radiology residents, professionals of allied sciences (including radiographers/radiological technologists, nuclear medicine physicians, medical physicists, and data scientists) & professionals of allied sciences in training residing within the boundaries of Europe

  • Reduced registration fees for ECR 1
  • Reduced fees for the European School of Radiology (ESOR) 2
  • Exclusive option to participate in the European Diploma. 3
  • Free electronic access to the journal European Radiology 4
  • Content e-mails for all ESR journals
  • Updates on offers & events through our newsletters

€ 11 /year

Yes! That is less than €1 per month.

Free membership

for radiologists, radiology residents or professionals of allied sciences engaged in practice, teaching or research residing outside Europe as well as individual qualified professionals with an interest in radiology and medical imaging who do not fulfil individual or all requirements for any other ESR membership category & former full members who have retired from all clinical practice
  • Reduced registration fees for ECR 1
  • Free electronic access to the journal European Radiology
  • Content e-mails for all 3 ESR journals 4
  • Updates on offers & events through our newsletters

€ 0

The best things in life are free.

ESR Friends

For students, company representatives or hospital managers etc.

  • Content e-mails for all 3 ESR journals 4
  • Updates on offers & events through our newsletters

€ 0

Friendship doesn’t cost a thing.

The membership type best fitting for you will be selected automatically during the application process.



Reduced registration fees for ECR 2024:
Provided that ESR 2023 membership is activated and approved by August 31, 2023 (excl. "Members in Training Offer").

Reduced registration fees for ECR 2025:
Provided that ESR 2024 membership is activated and approved by August 31, 2024.

Not all activities included
Examination based on the ESR European Training Curriculum (radiologists or radiology residents).
European Radiology, Insights into Imaging, European Radiology Experimental.