Polyp-D2ATL: Deep Domain-Adaptive Transfer Learning for Colorectal Polyp Classification under Label Distribution Shift

2026-06-12 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

Polyp-D2ATL introduces a novel deep domain-adaptive transfer learning framework designed for highly accurate colorectal polyp classification. This system specifically addresses critical challenges in real-world scenarios, including imbalanced data, label distribution shift, and the need for cross-modality generalization. The framework effectively predicts different classes of polyps according to the NICE classification. Extensive experiments on the PICCOLO validation and test sets demonstrate that Polyp-D2ATL significantly outperforms existing advanced models. It achieved an accuracy of 82.38%, a Macro-F1 of 77.49%, and a specificity of 87.47% on the validation set, showing consistent improvements and strong generalization capacity on the held-out test set, confirming its clinical applicability.

Key takeaway

For Computer Vision Engineers developing automated colorectal polyp classification systems, Polyp-D2ATL presents a robust framework addressing critical real-world challenges like imbalanced data and label distribution shift. You should investigate its deep domain-adaptive transfer learning approach to improve diagnostic accuracy and cross-modality generalization, especially when dealing with diverse clinical datasets. This could significantly enhance the reliability of your diagnostic tools.

Key insights

Polyp-D2ATL uses deep domain-adaptive transfer learning to classify colorectal polyps, effectively mitigating label distribution shift and data imbalance.

Principles

• Mitigating label distribution shift improves real-world applicability.
• Domain-adaptive transfer learning enhances cross-modality generalization.
• Specific training strategies are crucial for imbalanced medical data.

Method

Polyp-D2ATL employs a novel framework with a specific training strategy to classify polyps according to NICE classification, addressing imbalanced data and label distribution shift for improved prediction.

In practice

• Apply Polyp-D2ATL for automated polyp prediction.
• Use NICE classification for polyp categorization.
• Evaluate models using Macro-F1 and specificity.

Topics

• Colorectal Polyp Classification
• Deep Domain Adaptation
• Transfer Learning
• Label Distribution Shift
• Medical Imaging
• NICE Classification

Best for: AI Scientist, Computer Vision Engineer, Research Scientist

Related on AIssential

• DDF2Pol: A Dual-Domain Feature Fusion Network for PolSAR Image Classification — DDF2Pol fuses real and complex PolSAR features with attention for efficient, high-accuracy classification.
• RaLMPH: Reliability-aware Learning for Multi-Pathologist Harmonization in Whole-Slide Image Classification — RaLMPH uses a reliability field to harmonize multi-pathologist WSI annotations by identifying trustworthy local opinions.
• ASGNet: Adaptive Spectrum Guidance Network for Automatic Polyp Segmentation — Integrating spectral features with global attributes enhances polyp segmentation accuracy beyond spatial-only methods.
• Cut instance mixing: A domain-specific data augmentation method applied to gastrointestinal lesion detection — A novel domain-specific data augmentation method, Cut Instance Mixing (CIM), has been introduced to address the scarcity of well-annotated medical image datasets and the subtle appearance of gastroint…
• IEEE Transactions on Neural Networks and Learning Systems, Volume 37, Issue 3, March 2026 — The issue highlights advancements across neural networks, computer vision, and machine learning applications.
• Multi-Label Test-Time Adaptation with Bayesian Conditional Priors — BCP enhances multi-label VLM recognition under distribution shift by injecting label dependency via anchor-conditioned Bayesian refinement.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.