Multi-task SegFormer for knee ultrasound: Effusion ROI segmentation and patient-level classification of inflammatory arthritis

Şimşek Z., Tulum G., İnce M. D., Cuce F., Güneş E. Ç., Çaviş T., ...Daha Fazla

Pattern Recognition, cilt.180, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 180
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1016/j.patcog.2026.114358
  • Dergi Adı: Pattern Recognition
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, BIOSIS, Compendex, INSPEC, MLA - Modern Language Association Database, zbMATH, MLA International Bibliography, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
  • Anahtar Kelimeler: Effusion segmentation, Inflammatory arthritis, Knee ultrasound, Multi-task learning, SegFormer
  • Yıldız Teknik Üniversitesi Adresli: Evet

Özet

Knee ultrasound is a practical imaging modality for evaluating joint effusion, yet reliable delineation of the effusion region remains challenging due to low contrast, speckle noise, and indistinct boundaries. This study developed a multi-task SegFormer-based framework for effusion ROI segmentation and patient-level classification of inflammatory arthritis (IA) versus non-IA from knee ultrasound. The cohort comprised 65 individuals (26 IA, 39 non-IA) with 10 B-mode slices per patient. Effusion masks were generated using a semi-automated workflow and finalized by radiologist consensus. Evaluation was performed within a five-fold nested cross-validation framework with patient-level separation, comparing the proposed model with Mask R-CNN and YOLOv8-Seg. SegFormer achieved the best segmentation performance (Dice: 91.63 ± 1.25%, IoU: 85.57 ± 1.88%) and the most balanced classification profile (accuracy: 90.77 ± 6.44%, F1: 90.45 ± 6.48%). Mask R-CNN showed lower segmentation performance and a higher rate of missed IA cases, whereas YOLOv8-Seg identified all IA cases at the cost of markedly more false positives. These findings suggest that an ROI-aware multi-task SegFormer can effectively combine effusion segmentation with patient-level IA classification. Larger multicenter validation studies are warranted.