Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unsupervised Adaptation from FDG to PSMA PET/CT for 3D Lesion Detection under Label Shift

Xiaofeng Liu, Menghua Xia, Yanis Chemli, Georges El Fakhri, Chi Liu, Jinsong Ouyang

Abstract

In this work, we propose an unsupervised domain adaptation (UDA) framework for 3D volumetric lesion detection that adapts a detector trained on labeled FDG PET/CT to unlabeled PSMA PET/CT. Beyond covariate shift, cross tracer adaptation also exhibits label shift in both lesion size composition and the number of lesions per subject. We introduce self-training with two mechanisms that explicitly model and compensate for this label shift. First, we adaptively adjust the detection anchor shapes by re-estimating target domain box scales from selected pseudo labels and updating anchors with an exponential moving average. This increases positive anchor coverage for small PSMA lesions and stabilizes box regression. Second, instead of a fixed confidence threshold for pseudo-label selection, we allocate size bin-wise quotas according to the estimated target domain histogram over lesion volumes. The self-training alternates between supervised learning with prior-guided pseudo labeling on PSMA and supervised learning on labeled FDG. On AutoPET 2024, adapting from 501 labeled FDG studies to 369 $^{18}$F-PSMA studies, the proposed method improves both AP and FROC over the source-only baseline and conventional self-training without label-shift mitigation, indicating that modeling target lesion prevalence and size composition is an effective path to robust cross-tracer detection.

Unsupervised Adaptation from FDG to PSMA PET/CT for 3D Lesion Detection under Label Shift

Abstract

In this work, we propose an unsupervised domain adaptation (UDA) framework for 3D volumetric lesion detection that adapts a detector trained on labeled FDG PET/CT to unlabeled PSMA PET/CT. Beyond covariate shift, cross tracer adaptation also exhibits label shift in both lesion size composition and the number of lesions per subject. We introduce self-training with two mechanisms that explicitly model and compensate for this label shift. First, we adaptively adjust the detection anchor shapes by re-estimating target domain box scales from selected pseudo labels and updating anchors with an exponential moving average. This increases positive anchor coverage for small PSMA lesions and stabilizes box regression. Second, instead of a fixed confidence threshold for pseudo-label selection, we allocate size bin-wise quotas according to the estimated target domain histogram over lesion volumes. The self-training alternates between supervised learning with prior-guided pseudo labeling on PSMA and supervised learning on labeled FDG. On AutoPET 2024, adapting from 501 labeled FDG studies to 369 F-PSMA studies, the proposed method improves both AP and FROC over the source-only baseline and conventional self-training without label-shift mitigation, indicating that modeling target lesion prevalence and size composition is an effective path to robust cross-tracer detection.
Paper Structure (10 sections, 9 equations, 3 figures, 1 table)

This paper contains 10 sections, 9 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. 3D detector and source pretraining
  4. Epoch-wise anchor size adaptation
  5. Prior-guided pseudo-label selection
  6. Alternating optimization
  7. Experiments
  8. Discussion and Conclusion
  9. COMPLIANCE WITH ETHICAL STANDARDS
  10. ACKNOWLEDGMENTS

Figures (3)

  • Figure 1: (a-b) Examples of FDG and PSMA scans. (c) Lesion size distribution per subject for FDG vs. PSMA. On average, each PSMA subject contains more small lesions, whereas FDG has slightly more very large lesions ($\geq$150 cc). All volumes are resampled to $4\times4\times5$mm spacing.
  • Figure 2: Example lesion detection results on PSMA PET/CT.
  • Figure 3: Free-response ROC (FROC) curves on PSMA test set for the FDG-trained model (blue) and self-training UDA without (orange) or with (green) label shift correction.