A self-supervised pre-training method for lesion segmentation in oral potentially malignant disorders
Yuqi Cao, Miao Lu, Jiayuan Zhang, Rui Zhang, Linjun Shi, Hongxia Dan, Pingjie Huang, Guangxin Zhang, Qianming Chen · 2026
Read A self-supervised pre-training method for lesion segmentation in oral potentially malignant disorders by Yuqi Cao, Miao Lu, Jiayuan Zhang, Rui Zhang, Linjun Shi, Hongxia Dan, Pingjie Huang, Guangxin Zhang, Qianming Chen (2026) 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/PMC13479575.txt.
Description
Yuqi Cao and colleagues present a 2026 computer vision study on outlining lesions in photographs of oral potentially malignant disorders, mouth changes that can precede cancer. Training segmentation models needs expert-drawn outlines, which are scarce where specialists are few. The team built a self-supervised pre-training method that learns image features from unlabeled photos first.
Their internal set held 3,417 photographs, with two outside hospital datasets reserved for validation. Patient-level splits prevented the same person's images from leaking across training and testing. The model, a multiscale attention design paired with self-generated masks and teacher-student consistency rules, reached about 84 percent on the Dice overlap score. The authors aim to carry specialist-level outlining into clinics that lack specialists.
Catalog record
| CREATOR | Yuqi Cao, Miao Lu, Jiayuan Zhang, Rui Zhang, Linjun Shi, Hongxia Dan, Pingjie Huang, Guangxin Zhang, Qianming Chen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56193 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |