From Fully-Automatic to Efficient Interaction: A Lightweight Prompting Framework for Jaw Cyst Segmentation

Faint, interrupted boundaries make jaw cysts difficult to delineate in medical images. Clinicians can help identify the lesion, yet a fully automatic segmentation model has no direct means of using their assessment to guide an individual mask. We developed PromptFormer to accept bounding box prompts through a compact medical segmentation backbone. Its Plug-in Prompt Process Module (PPPM) couples that backbone to the prompt encoder and mask decoder of the Segment Anything Model (SAM); SAM’s image encoder is omitted. The backbone is supplemented by a Frequency-domain Global Perception Module (FGPM) for global structure and a Multi-scale Detail Refinement Module (MDRM) for spatial detail. Tests used a clinically collected jaw cyst dataset, BUSI, and ISIC 2017. Across the three datasets, the highest Intersection over Union (IoU) among the compared models was obtained by PromptFormer. Its jaw cyst Dice score was 93.79%. Relative to the SAM-based comparators in the efficiency test, fewer than one seventh as many floating-point operations were needed, and inference was approximately 2.7 to 8.6 times as fast. These computational savings, together with the prompt interface, justify further evaluation on clinical devices with limited resources.

Authors

Institutions

Publication Details

Journal
Transactions on Artificial Intelligence
Published
2026-11-17
DOI
https://doi.org/10.53941/tai.2026.100015
Primary Topic
Medical Image Segmentation Techniques
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

From Fully-Automatic to Efficient Interaction: A Lightweight Prompting Framework for Jaw Cyst Segmentation

Tinglong Tang, Yalin Wang, Xiaotian Pan, Zhe Rao et al.
4 links
Transactions on Artificial Intelligence
Medical Image Segmentation Techniques
article

From Fully-Automatic to Efficient Interaction: A Lightweight Prompting Framework for Jaw Cyst Segmentation

Tinglong Tang, Yalin Wang, Xiaotian Pan, Zhe Rao, Chao Li, Zhenyang Wang
article en

Abstract

Faint, interrupted boundaries make jaw cysts difficult to delineate in medical images. Clinicians can help identify the lesion, yet a fully automatic segmentation model has no direct means of using their assessment to guide an individual mask. We developed PromptFormer to accept bounding box prompts through a compact medical segmentation backbone. Its Plug-in Prompt Process Module (PPPM) couples that backbone to the prompt encoder and mask decoder of the Segment Anything Model (SAM); SAM’s image encoder is omitted. The backbone is supplemented by a Frequency-domain Global Perception Module (FGPM) for global structure and a Multi-scale Detail Refinement Module (MDRM) for spatial detail. Tests used a clinically collected jaw cyst dataset, BUSI, and ISIC 2017. Across the three datasets, the highest Intersection over Union (IoU) among the compared models was obtained by PromptFormer. Its jaw cyst Dice score was 93.79%. Relative to the SAM-based comparators in the efficiency test, fewer than one seventh as many floating-point operations were needed, and inference was approximately 2.7 to 8.6 times as fast. These computational savings, together with the prompt interface, justify further evaluation on clinical devices with limited resources.

Transactions on Artificial IntelligenceVol. 2(1)
China Three Gorges University (CN), Zhijiang College of Zhejiang University of Technology (CN), Hangzhou Dianzi University (CN), Zhejiang University of Technology (CN), Lanzhou University (CN)
National Natural Science Foundation of China, Natural Science Foundation of Zhejiang Province
Medical Image Segmentation Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon