SynPAIN
收藏资源简介:
SynPAIN是一个大规模的合成数据集,包含10710张面部表情图像(5355对中性/表情图像),跨越五个种族/民族,代表两个年龄组(年轻:20-35岁,老年:75岁以上),以及两个性别。使用商业生成式AI工具,我们创建了人口统计上平衡的合成身份,具有临床上有意义的疼痛表情。我们的验证表明,合成疼痛表情表现出预期的疼痛模式,使用基于面部动作单元分析的临床验证的疼痛评估工具,得分显著高于中性和非疼痛表情。我们通过实验证明SynPAIN在识别现有疼痛检测模型中的算法偏差方面的实用性。通过全面的偏差评估,我们揭示了在人口统计特征方面存在重大性能差异。这些性能差异以前在较小、不太多样化的数据集中是无法检测到的。此外,我们还证明,年龄匹配的合成数据增强提高了在真实临床数据上的疼痛检测性能,平均精确度提高了7.0%。SynPAIN通过提供第一个公开可用的、人口统计上多样化的合成数据集,专门为老年人疼痛检测而设计,同时建立了一个衡量和减轻算法偏差的框架,从而解决了疼痛评估研究中的关键差距。数据集可在SynPAIN获取。
SynPAIN is a large-scale synthetic dataset containing 10,710 facial expression images (5,355 pairs of neutral/expression images) spanning five racial/ethnic groups, covering two age groups (young: 20–35 years old, elderly: 75 years and older) and two genders. Using commercial generative AI tools, we developed demographically balanced synthetic identities with clinically meaningful pain expressions. Our validation experiments show that the synthetic pain expressions exhibit expected pain patterns, with scores significantly higher than those of neutral and non-pain expressions as measured by clinically validated pain assessment tools based on facial action unit analysis. We experimentally demonstrate the utility of SynPAIN in identifying algorithmic bias in existing pain detection models. Through comprehensive bias assessments, we reveal significant performance disparities across demographic characteristics, which were previously undetectable in smaller, less diverse datasets. Furthermore, we show that age-matched synthetic data augmentation improves pain detection performance on real clinical data, with a 7.0% increase in mean average precision. SynPAIN addresses critical gaps in pain assessment research by providing the first publicly available, demographically diverse synthetic dataset specifically designed for pain detection in the elderly, while establishing a framework for measuring and mitigating algorithmic bias. The dataset is accessible at SynPAIN.

- 1SynPAIN: A Synthetic Dataset of Pain and Non-Pain Facial Expressions多伦多大学计算机科学系, KITE Research Institute, Toronto Rehabilitation Institute, University Health Network, Institute of Biomedical Engineering, University of Toronto, Vector Institute, 多伦多, 加拿大; 里贾纳大学, 里贾纳, 加拿大 · 2025年



