隨著運算能力及硬體資源的提升，基礎模型在醫學影像分析的領域獲得了顯著的進展。這些模型使用大量醫學影像的資料進行預訓練，在各種臨床影像任務中均展現出專家級的表現。然而，由於基礎模型會在這樣高風險的臨床上進行應用，這些透過資料驅動的模型中可能存在的潛在偏差問題而逐漸引起關注。先前已有研究表明，基礎模型會受資料不平衡的影響，尤其是在代表性不足的性別少數或種族少數的患者中，易產生公平性問題。公開數據集中缺乏來自這些群體的樣本，可能導致模型受到敏感屬性的影響而忽略核心的分類特徵。因此，儘管這些模型可能表現出良好的平均預測能力，但少數族群的診斷準確性產生偏誤，進而對醫療的公平性造成負面影響。
為了應對這一挑戰，本研究旨在評估並消弭基礎模型在病理切片影像中的癌症診斷偏誤，同時盡可能地不犧牲模型的預測能力。我們提出了一種簡單而有效的基於公平的不平衡校正微調策略，透過對模型進行前幾層參數的凍結保留核心判別的特徵，加入損失函數引導公平的預測，來提升模型預測的公平性。我們選用了腫瘤偵測、癌症分類、存活分析和遺傳變異分析這幾種下游任務進行實驗。結果顯示，與未微調之基礎模型相比，我們的方法在不犧牲平均準確性的情況下，顯著提高了多項任務的公平性指標，證明此方法能部分解決資料不平衡導致的模型偏誤問題。本研究的研究結果為臨床應用提出了更好的解決方案，並有助於人工智慧在醫學領域更公平的發展。

With the advancement of computation power, foundation models have gained significant traction in medical image analysis. These models, pretrained on medical domain data, exhibit expert-level performance across various clinical imaging tasks. However, concerns have arisen about potential biases embedded within these data-driven systems as they are utilized in high-stakes clinical applications. Prior studies indicate that foundation models often struggle to address data imbalance issues, particularly for underrepresented sexual or racial demographic subgroups. The scarcity of data from these populations in publicly available datasets can result in predictions being influenced by inherent factors rather than core target features. Consequently, while these models may demonstrate strong average predictive ability, diagnostic accuracy for minority subgroups can be compromised, negatively impacting healthcare equity.
To address this challenge, this study aims to evaluate and enhance the fairness of foundation models for histopathology images without degrading the predictive performance. We propose a simple yet effective strategy, Fairness-Aware Imbalance Rectification Tuning (FAIR-Tuning), selectively regularizing the model to guide equitable predictions while preserving the core discriminative representations. Experiments are conducted on tumor detection, cancer classification, survival prediction, and genetic mutation tasks. Quantitative results demonstrate that our approach achieves significantly improved subgroup fairness metrics compared to baseline models without sacrificing average accuracy, suggesting its potential to address data imbalance issues and help reduce population disparities. Our findings provide insights into developing more inclusive AI systems for clinical decision support.

[1] Ministry of Health and Welfare. 112年國人死因統計結果, 2024, June. Accessed: 2024-07-02.
[2] Zizhao Zhang, Pingjun Chen, M. McGough, F. Xing, Chunbao Wang, M. Bui, Yuanpu Xie, Manish Sapkota, Lei Cui, J. Dhillon, Nazeel Ahmad, F. Khalil, S. Dickinson, Xiaoshuang Shi, Fujun Liu, H. Su, Jinzheng Cai, and Lin Yang. Pathologist-level interpretable whole-slide cancer diagnosis with deep learning. Nature Machine Intelligence, 1:236–245, 2019.
[3] Ian A. Cree, Ian A. Cree, Z. Deans, M. Ligtenberg, N. Normanno, A. Edsjo, E. Rouleau, F. Sol ́e, E. Thunnissen, W. Timens, E. Schuuring, E. Dequeker, S. Murray, M. Dietel, P. Groenen, and J. H. Krieken. Guidance for laboratories performing molecular pathology for cancer patients. Journal of Clinical Pathology, 67:923 – 931, 2014.
[4] H. Irshad, A. Veillard, L. Roux, and Daniel Racoceanu. Methods for nuclei detection, segmentation, and classification in digital histopathology: A review—current status and future potential. IEEE Reviews in Biomedical Engineering, 7:97–114, 2014.
[5] G. Pena and J. S. Andrade-Filho. How does a pathologist make a diagnosis? Archives of pathology & laboratory medicine, 133 1:124–32, 2009.
[6] Alexander Rakhlin, Alexey Shvets, Vladimir Iglovikov, and Alexandr A Kalinin. Deep convolutional neural networks for breast cancer histology image analysis. In Image Analysis and Recognition: 15th International Conference, ICIAR 2018, P ́ovoa de Varzim, Portugal, June 27–29, 2018, Proceedings 15, pages 737–744. Springer, 2018.
[7] Manisha Saini and Seba Susan. Vggin-net: Deep transfer network for imbalanced breast cancer dataset. IEEE/ACM Transactions on Computational Biology and Bioinformatics, 20(1):752–762, 2022.
[8] Prakash Chandra Chhipa, Richa Upadhyay, Gustav Grund Pihlgren, Rajkumar Saini, Seiichi Uchida, and Marcus Liwicki. Magnification prior: a self-supervised method for learning representations on breast cancer histopathological images. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 2717–2727, 2023.
[9] Bobby Azad, Reza Azad, Sania Eskandari, Afshin Bozorgpour, Amirhossein Kazerouni, Islem Rekik, and Dorit Merhof. Foundational models in medical imaging: A comprehensive survey and future vision. arXiv preprint arXiv:2310.18689, 2023.
[10] Laila Rasmy, Yang Xiang, Ziqian Xie, Cui Tao, and Degui Zhi. Med-bert: pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction. NPJ digital medicine, 4(1):86, 2021.
[11] Jun Ma, Yuting He, Feifei Li, Lin Han, Chenyu You, and Bo Wang. Segment anything in medical images. Nature Communications, 15(1):654, 2024.
[12] Laleh Seyyed-Kalantari, Haoran Zhang, Matthew BA McDermott, Irene Y Chen, and Marzyeh Ghassemi. Underdiagnosis bias of artificial intelligence algorithms applied to chest radiographs in under-served patient populations. Nature medicine, 27(12):2176–2182, 2021.
[13] Kristine Bærøe, Torbjørn Gundersen, Edmund Henden, and Kjetil Rommetveit. Can medical algorithms be fair? three ethical quandaries and one dilemma. BMJ Health & Care Informatics, 29(1), 2022.
[14] Yongshuo Zong, Yongxin Yang, and Timothy Hospedales. Medfair: Benchmarking fairness for medical imaging. arXiv preprint arXiv:2210.01725, 2022.
[15] Shaoting Zhang and Dimitris N. Metaxas. On the challenges and perspectives of foundation models for medical image analysis. Medical image analysis, 91:102996, 2023.
[16] Yukun Zhou, Mark A Chia, Siegfried K Wagner, Murat S Ayhan, Dominic J Williamson, Robbert R Struyven, Timing Liu, Moucheng Xu, Mateo G Lozano, Peter Woodward-Court, et al. A foundation model for generalizable disease detection from retinal images. Nature, 622(7981):156–163, 2023.
[17] Yang Yuan. On the power of foundation models. In International Conference on Machine Learning, pages 40519–40530. PMLR, 2023.
[18] Dequan Wang, Xiaosong Wang, Lilong Wang, Mengzhang Li, Qian Da, Xiaoqiang Liu, Xiangyu Gao, Jun Shen, Junjun He, Tian Shen, et al. A real-world dataset and benchmark for foundation model adaptation in medical image classification. Scientific Data, 10(1):574, 2023.
[19] Xiaosong Wang, Yifan Peng, Le Lu, Zhiyong Lu, M. Bagheri, and R. Summers. Chestx-ray: Hospital-scale chest x-ray database and benchmarks on weakly supervised classification and localization of common thorax diseases. Pages 369–392, 2019.
[20] Zelong Liu, Andrew Tieu, Nikhil Patel, Alexander Zhou, George Soultanidis, Zahi A Fayad, Timothy Deyer, and Xueyan Mei. Vision-mae: A foundation model for medical image segmentation and classification. arXiv preprint arXiv:2402.01034, 2024.
[21] Xiyue Wang, Sen Yang, Jun Zhang, Minghui Wang, Jing Zhang, Junzhou Huang, Wei Yang, and Xiao Han. Transpath: Transformer-based self-supervised learning for histopathological image classification. In Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part VIII 24, pages 186–195. Springer, 2021.
[22] Simon Caton and Christian Haas. Fairness in machine learning: A survey. ACM Computing Surveys, 56(7):1–38, 2024.
[23] Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram Galstyan. A survey on bias and fairness in machine learning. ACM computing surveys (CSUR), 54(6):1–35, 2021.
[24] Laleh Seyyed-Kalantari, Guanxiong Liu, Matthew McDermott, Irene Y Chen, and Marzyeh Ghassemi. Chexclusion: Fairness gaps in deep chest x-ray classifiers. In BIOCOMPUTING 2021: proceedings of the Pacific symposium, pages 232–243. World Scientific, 2020.
[25] Ziad Obermeyer and Sendhil Mullainathan. Dissecting racial bias in an algorithm that guides health decisions for 70 million people. In Proceedings of the conference on fairness, accountability, and transparency, pages 89–89, 2019.
[26] Philip Adler, Casey Falk, Sorelle A. Friedler, Tionney Nix, Gabriel Rybeck, C. Scheidegger, Brandon Smith, and Suresh Venkatasubramanian. Auditing black-box models for indirect influence. Knowledge and Information Systems, 54:95 –122, 2016.
[27] Wenyue Hua, Yingqiang Ge, Shuyuan Xu, Jianchao Ji, and Yongfeng Zhang. Up5: Unbiased foundation model for fairness-aware recommendation. arXiv preprint arXiv:2305.12090, 2023.
[28] Agostina J Larrazabal, Nicol ́as Nieto, Victoria Peterson, Diego H Milone, and Enzo Ferrante. Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. Proceedings of the National Academy of Sciences, 117(23):12592–12594, 2020.
[29] Aayush Gupta, Ayush Jaiswal, Yuehua Wu, Vivek Yadav, and P. Natarajan. Adversarial mask generation for preserving visual privacy. 2021 16th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2021), pages 1–5, 2021.
[30] Chaoning Zhang, Philipp Benz, Dawit Mureja Argaw, Seokju Lee, Junsik Kim, Francois Rameau, Jean-Charles Bazin, and In So Kweon. Resnet or densenet? introducing dense shortcuts to resnet. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 3550–3559, 2021.
[31] Christopher T. H. Teo, Milad Abdollahzadeh, and Ngai-Man Cheung. Fair generative models via transfer learning. ArXiv, abs/2212.00926, 2022.
[32] Mahdi Erfanian, HV Jagadish, and Abolfazl Asudeh. Chameleon: Foundation models for fairness-aware multi-modal data augmentation to enhance coverage of minorities. arXiv preprint arXiv:2402.01071, 2024.
[33] Changjian Shui, Justin Szeto, Raghav Mehta, Douglas L Arnold, and Tal Arbel. Mitigating calibration bias without fixed attribute grouping for improved fairness in medical imaging analysis. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 189–198. Springer, 2023.
[34] Zhuoyan Xu, Zhenmei Shi, Junyi Wei, Fangzhou Mu, Yin Li, and Yingyu Liang. Towards few-shot adaptation of foundation models via multitask finetuning. arXiv preprint arXiv:2402.15017, 2024.
[35] Yuzhen Mao, Zhun Deng, Huaxiu Yao, Ting Ye, Kenji Kawaguchi, and James Zou. Last-layer fairness fine-tuning is simple and effective for neural networks. arXiv preprint arXiv:2304.03935, 2023.
[36] Zeyu Wang, Klint Qinami, Ioannis Christos Karakozis, Kyle Genova, Prem Nair, Kenji Hata, and Olga Russakovsky. Towards fairness in visual recognition: Effective strategies for bias mitigation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8919–8928, 2020.
[37] Bashir Sadeghi and Vishnu Naresh Boddeti. Imparting fairness to pre-trained biased representations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 16–17, 2020.
[38] Zhi Huang, Federico Bianchi, Mert Yuksekgonul, Thomas J Montine, and James Zou. A visual–language foundation model for pathology image analysis using medical twitter. Nature medicine, 29(9):2307–2316, 2023.
[39] Eugene Vorontsov, Alican Bozkurt, Adam Casson, George Shaikovski, Michal Zelechowski, Siqi Liu, Philippe Mathieu, Alexander van Eck, Donghun Lee, Julian Viret, et al. Virchow: a million-slide digital pathology foundation model. arXiv preprint arXiv:2309.07778, 2023.
[40] Hanwen Xu, Naoto Usuyama, Jaspreet Bagga, Sheng Zhang, Rajesh Rao, Tristan Naumann, Cliff Wong, Zelalem Gero, Javier Gonz ́alez, Yu Gu, et al. A whole-slide foundation model for digital pathology from real-world data. Nature, pages 1–8, 2024.
[41] Xiyue Wang, Junhan Zhao, Eliana Marostica, Wei Yuan, Jietian Jin, Jiayu Zhang, Ruijiang Li, Hongping Tang, Kanran Wang, Yu Li, Fang Wang, Yulong Peng, Junyou Zhu, Jing Zhang, Christopher R. Jackson, Jun Zhang, Deborah Dillon, Nancy U. Lin, Lynette Sholl, Thomas Denize, David Meredith, Keith L. Ligon, Sabina Signoretti, Shuji Ogino, Jeffrey A. Golden, MacLean P. Nasrallah, Xiao Han, Sen Yang, and Kun-Hsing Yu. A pathology foundation model for cancer diagnosis, molecular profile evaluation, and prognostic prediction. Nature, 2024. (in press).
[42] Peizhao Li, Han Zhao, and Hongfu Liu. Deep fair clustering for visual learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9070–9079, 2020.
[43] Francis McKay, Bethany J Williams, Graham Prestwich, Daljeet Bansal, Nina Hallowell, and Darren Treanor. The ethical challenges of artificial intelligence-driven digital pathology. The journal of pathology: clinical research, 8(3):209–216, 2022.
[44] Dohyung Kim, Sungho Park, Sunhee Hwang, and Hyeran Byun. Fair classification by loss balancing via fairness-aware batch sampling. Neurocomputing, 518:231–241, 2023.
[45] Katarzyna Tomczak, Patrycja Czerwi ́nska, and Maciej Wiznerowicz. Review the cancer genome atlas (tcga): an immeasurable source of knowledge. Contemporary Oncology/Wsp ́o lczesna Onkologia, 2015(1):68–77, 2015.