Adaptive Voxel-Weighted Loss Using L1 Norms in Deep Neural Networks for Detection and Segmentation of Prostate Cancer Lesions in PET/CT Images
Obed K. Dzikunu, Shadab Ahamed, Amirhossein Toosi, Sara Harsini, François Bénard, Xiaoxiao Li, Arman Rahmim · 2025
Read Adaptive Voxel-Weighted Loss Using L1 Norms in Deep Neural Networks for Detection and Segmentation of Prostate Cancer Lesions in PET/CT Images by Obed K. Dzikunu, Shadab Ahamed, Amirhossein Toosi, Sara Harsini, François Bénard, Xiaoxiao Li, Arman Rahmim (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-2502-02756.txt.
Description
Obed Dzikunu and co-authors sharpen AI spotting of prostate cancer hot spots in scans. Repeat cancer can hide in lesions of wild size and brightness, amid mostly healthy tissue. Standard training losses let easy background points rule the learning, or let wild outliers yank it off course. Their new loss weights each point by how hard it is to class.
The weighting uses a plain norm to even out the push of each point. Four network designs trained on hundreds of scans all gained from the swap. Detection scores rose, and false alarms fell, across lesion sizes. A large medical foundation model, fine tuned with the loss, gained too. Code is public.
Catalog record
| CREATOR | Obed K. Dzikunu, Shadab Ahamed, Amirhossein Toosi, Sara Harsini, François Bénard, Xiaoxiao Li, Arman Rahmim |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-02756