Exploring connections of spectral analysis and transfer learning in medical imaging
Not on the map. Why?
Abstract (arXiv)
In this paper, we use spectral analysis to investigate transfer learning and study model sensitivity to frequency shortcuts in medical imaging. By analyzing the power spectrum density of both pre-trained and fine-tuned model gradients, as well as artificially generated frequency shortcuts, we observe notable differences in learning priorities between models pre-trained on natural vs medical images, which generally persist during fine-tuning. We find that when a model's learning priority aligns with the power spectrum density of an artifact, it results in overfitting to that artifact. Based on these observations, we show that source data editing can alter the model's resistance to shortcut learning.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 18 of its 31 references that are in the map. First posted on arXiv 16 Jul 2024.
Builds on
18 of 31 references.
References · 18
- 2015Deep Residual Learning for Image Recognitioncited by 52,096
- 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
- 2020Shortcut Learning in Deep Neural Networkscited by 1,969
- 2018On the Spectral Bias of Neural Networkscited by 1,341
- 2019Transfusion: Understanding Transfer Learning for Medical Imagingcited by 368
- 2019High Frequency Component Helps Explain the Generalization of Convolutional Neural Networkscited by 310
- 2021A Fourier-based Framework for Domain Generalizationcited by 290
- 2019A Fourier Perspective on Model Robustness in Computer Visioncited by 288
- 2021Amplitude-Phase Recombination: Rethinking Robustness of Convolutional Neural Networks in Frequency Domaincited by 71
- 2021Spectral Bias in Practice: The Role of Function Frequency in Generalizationcited by 37
- 2022Self-supervised Learning from 100 Million Medical Imagescited by 31
- 2023What do neural networks learn in image classification? A frequency shortcut perspectivecited by 28
6 more
- 2018Cats or CAT scans: transfer learning from natural or medical image source datasets?cited by 14
- 2023HybridAugment++: Unified Frequency Spectra Perturbations for Model Robustnesscited by 12
- 2024Source Matters: Source Dataset Impact on Model Robustness in Medical Imagingcited by 10
- 2022Investigating and Explaining the Frequency Bias in Image Classificationcited by 8
- 2023Robustmix: Improving Robustness by Regularizing the Frequency Bias of Deep Netscited by 4
- 2023DFM-X: Augmentation by Leveraging Prior Knowledge of Shortcut Learningcited by 4
Built on by
Citing papers per year: 2025: 1.
1 paper.