Abstract:This manuscript presents an image segmentation algorithm developed for the Cross-Organ and Cross-Scanner Adenocarcinoma Segmentation (COSAS 2024) challenge. We adopted an organ-stratified and scanner-stratified approach to train multiple Upernet-based segmentation models and subsequently ensembled the results. Despite the challenges posed by the varying tumor characteristics across different organs and the differing imaging conditions of various scanners, our method achieved a final test score of 0.7643 for Task 1 and 0.8354 for Task 2. These results demonstrate the adaptability and efficacy of our approach across diverse conditions. Our model's ability to generalize across various datasets underscores its potential for real-world applications.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the domain adaptation problem in adenocarcinoma segmentation across organs and scanners. Specifically, the author focuses on the problem of insufficient generalization ability of existing segmentation algorithms in practical applications due to image variations (i.e., domain shift) brought by different organs and different scanners in digital pathology images. ### Problem Background 1. **Domain Shift**: - The diversity of digital pathology images poses significant challenges. These images may come from different organs, use different tissue preparation techniques, and different image acquisition methods. - These differences lead to the so - called "domain shift", making it difficult for the model to maintain consistent performance under different conditions. 2. **Limitations of Existing Methods**: - Current algorithms are usually optimized for specific types of cancer but often overlook the unique characteristics of different organs and the image differences generated by different scanners. - This limitation restricts the application of the model in the real - world clinical environment. ### Research Objectives To address the above problems, this paper proposes a new segmentation algorithm, aiming to: - **Improve the Generalization Ability of the Model**: By adopting the methods of "organ - stratified training" and "scanner - stratified training", ensure that the model can perform well on data from different organs and different scanners. - **Evaluate Domain Adaptation Techniques**: Utilize the extensive data set provided by the COSAS 2024 challenge to evaluate the effectiveness of the proposed domain adaptation techniques. - **Achieve Better Clinical Application Potential**: By improving the generalization ability and robustness of the model, make it more suitable for actual clinical scenarios. ### Solution Overview The author adopts the following strategies to solve these problems: 1. **Organ - stratified Training**: - For Task 1, use three - fold cross - validation. Each time, take the data of one organ type as the validation set, and the data of the other two organ types as the training set. - This method ensures that the organ types in each training and validation set are completely separated, thereby better evaluating the generalization ability of the model on different organs. 2. **Scanner - stratified Training**: - For Task 2, also use three - fold cross - validation, but this time stratify based on scanners. Each time, take the data of one scanner as the validation set, and the data of the other two scanners as the training set. - This method helps to evaluate the generalization ability of the model under different scanner conditions. 3. **Model Ensemble**: - Use two methods, hard voting and probability averaging, to integrate the results of multiple models to further improve the accuracy of prediction. Through these methods, the author has successfully improved the segmentation performance of the model under different organ and scanner conditions and achieved excellent results in the COSAS 2024 challenge. ### Results and Conclusions Finally, this method achieved good results in the COSAS 2024 challenge, especially in handling the variations between different scanners. The author believes that in the future, the performance of the model can be further improved by introducing more domain adaptation techniques.

Domain-stratified Training for Cross-organ and Cross-scanner Adenocarcinoma Segmentation in the COSAS 2024 Challenge

Comprehensive and Clinically Accurate Head and Neck Cancer Organs-at-risk Delineation on a Multi-Institutional Study

Comprehensive and Clinically Accurate Head and Neck Organs at Risk Delineation Via Stratified Deep Learning: A Large-scale Multi-Institutional Study

Domain and Content Adaptive Convolutions for Cross-Domain Adenocarcinoma Segmentation

Cross-Task Pretraining for Cross-Organ Cross-Scanner Adenocarcinoma Segmentation

Cross-Organ and Cross-Scanner Adenocarcinoma Segmentation using Rein to Fine-tune Vision Foundation Models

Adenocarcinoma Segmentation Using Pre-trained Swin-UNet with Parallel Cross-Attention for Multi-Domain Imaging

Understanding Stain Separation Improves Cross-Scanner Adenocarcinoma Segmentation with Joint Multi-Task Learning

Cross-Organ Domain Adaptive Neural Network for Pancreatic Endoscopic Ultrasound Image Segmentation

Automatic Organ and Pan-cancer Segmentation in Abdomen CT: the FLARE 2023 Challenge

Multi-organ segmentation of abdominal structures from non-contrast and contrast enhanced CT images

Cross-Model Attention-Guided Tumor Segmentation for 3D Automated Breast Ultrasound (ABUS) Images

Organ at Risk Segmentation in Head and Neck CT Images Using a Two-Stage Segmentation Framework Based on 3D U-Net

Segmentation of the prostate and organs at risk in male pelvic CT images using deep learning

PheoSeg: A 3D transfer learning framework for accurate abdominal CT pheochromocytoma segmentation and surgical grade prediction

AG-CRC: Anatomy-Guided Colorectal Cancer Segmentation in CT with Imperfect Anatomical Knowledge

CT Male Pelvic Organ Segmentation Using Fully Convolutional Networks with Boundary Sensitive Representation

Cascaded SE-ResUnet for segmentation of thoracic organs at risk

Multiscale Local Enhancement Deep Convolutional Networks for the Automated 3D Segmentation of Gross Tumor Volumes in Nasopharyngeal Carcinoma: A Multi-Institutional Dataset Study

A 2D dilated residual U-Net for multi-organ segmentation in thoracic CT

Two-Stage Hybrid Supervision Framework for Fast, Low-resource, and Accurate Organ and Pan-cancer Segmentation in Abdomen CT