Anisotropic Fourier Features for Positional Encoding in Medical Imaging
Nabil Jabareen, Dongsheng Yuan, Dingming Liu, Foo-Wei Ten, Sören Lukassen · 2025
Read Anisotropic Fourier Features for Positional Encoding in Medical Imaging by Nabil Jabareen, Dongsheng Yuan, Dingming Liu, Foo-Wei Ten, Sören Lukassen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-02488.txt.
Description
Nabil Jabareen and four co-authors wrote a 2025 medical imaging paper. It covers positional encodings. These codes tell a transformer model where each image patch sits. Medical scans are often anisotropic. That means the spacing differs by direction. Standard codes ignore that fact.
Sine-based codes fail to keep distances true in higher dimensions. Isotropic Fourier codes fix distances but assume equal spacing in all directions. The new code adapts to direction, class, and domain. Tests cover chest X-ray labelling, organ sorting in CT scans, and heart pump readings from sound-based video. The choice of code changed results a lot. The new code beat top rivals in each test with uneven spacing. The authors say the code should fit the shape of the scan.
Catalog record
| CREATOR | Nabil Jabareen, Dongsheng Yuan, Dingming Liu, Foo-Wei Ten, Sören Lukassen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-02488