Comparison of the Effectiveness of Semi-Supervised and Self-Supervised Learning in Low Data Regimes in Medical Image Analysis Tibbi Görüntü Analizi Düşük Veri Rejimlerinde Yari Denetimli ve Kendi Kendini Denetleyen Ö?grenmenin Etkinli?ginin Karşilaştirilmasi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636791
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Biomedical Images, ConvNeXt, Self-Supervised Learning, Semi-Supervised Learning, TissueMNIST
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
This study compares supervised, semi-supervised (EMA-FixMatch), and self-supervised (SimSiam) learning approaches on the TissueMNIST dataset to address the labeled data scarcity in histopathological image analysis. The ConvNeXt-Tiny architecture was used as the common backbone in all experiments; the models were tested under label ratios of 5%, 10%, and 100%. The results show that supervised learning achieved the highest performance with 71.4% accuracy in the fully labeled scenario; in the limited-label scenarios, the EMA-FixMatch method outperformed the SimSiam-based approach (65.1% with 10% labels; 59.5% with 5% labels). The results highlight the effectiveness of semi-supervised learning under data scarcity and the reliability of supervised transfer learning when sufficient data is available.