A Two-Phase Deep Learning Approach for Detecting Laterally Spreading Tumors Utilizing Self-Supervised Learning and Few-Shot Classification Techniques - Takeaways - MDSpire

A Two-Phase Deep Learning Approach for Detecting Laterally Spreading Tumors Utilizing Self-Supervised Learning and Few-Shot Classification Techniques

  • By

  • Menghui Wang

  • Zhanpeng Shi

  • Yiwen Wang

  • Jie Lu

  • March 1, 2026

  • 0 min

Share

  • 1

    Laterally spreading tumors (LSTs) are significant precursors to colorectal cancer, with a prevalence of 0.8% in the colonoscopy population.

  • 2

    Deep learning techniques can enhance the detection of LSTs, which are challenging to identify due to their distinct features and low prevalence.

  • 3

    The proposed two-phase AI framework utilizes self-supervised learning and few-shot classification to improve LST detection with minimal labeled data.

  • 4

    A labeled dataset of 4000 images was created from 150,168 colonoscopy images, enabling effective training and evaluation of the AI model.

  • 5

    The study demonstrates that clinically relevant performance can be achieved with only 2799 labeled images, significantly reducing data requirements.

Original Source(s)

A Two-Phase Deep Learning Approach for Detecting Laterally Spreading Tumors Utilizing Self-Supervised Learning and Few-Shot Classification Techniques

  • by Menghui Wang, Zhanpeng Shi, Yiwen Wang, Jie Lu

  • March 1, 2026

Related Content

Comparison of the mixed approach and medial approach in laparoscopic right hemicolectomy for right colon cancer: a retrospective study

Case Report: Preoperative treatment of portal hypertension by splenic artery embolization for safe major hepatectomy: experience in three patients

Susceptibility to inflammatory bowel diseases promotes invasive carcinomas in a murine model of ATF6-driven colon cancer