Abstract
Face recognition applications have grown in parallel with the size of datasets, complexity of deep learning models and computational power. However, while deep learning models evolve to become more capable and computational power keeps increasing, the datasets available are being retracted and removed from public access. Privacy and ethical concerns are relevant topics within these domains. Through generative artificial intelligence, researchers have put efforts into the development of completely synthetic datasets that can be used to train face recognition systems. Nonetheless, the recent advances have not been sufficient to achieve performance comparable to the state-of-the-art models trained on real data. To study the drift between the performance of models trained on real and synthetic datasets, we leverage a massive attribute classifier (MAC) to create annotations for four datasets: two real and two synthetic. From these annotations, we conduct studies on the distribution of each attribute within all four datasets. Additionally, we further inspect the differences between real and synthetic datasets on the attribute set. When comparing through the Kullback-Leibler divergence we have found differences between real and synthetic samples. Interestingly enough, we have verified that while real samples suffice to explain the synthetic distribution, the opposite could not be further from being true.

Abstract
This study investigates the effects of occlusions on the fairness of face recognition systems, particularly focusing on demographic biases. Using the Racial Faces in the Wild (RFW) dataset and synthetically added realistic occlusions, we evaluate their effect on the performance of face recognition models trained on the BUPT-Balanced and BUPT-GlobalFace datasets. We note increases in the dispersion of FMR, FNMR, and accuracy alongside decreases in fairness according to Equalized Odds, Demographic Parity, STD of Accuracy, and Fairness Discrepancy Rate. Additionally, we utilize a pixel attribution method to understand the importance of occlusions in model predictions, proposing a new metric, Face Occlusion Impact Ratio (FOIR), that quantifies the extent to which occlusions affect model performance across different demographic groups. Our results indicate that occlusions exacerbate existing demographic biases, with models placing higher importance on occlusions in an unequal fashion across demographics. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Abstract

Abstract

Abstract
TPS621 Background: Breast cancer treatments often pose challenges in balancing efficacy with quality of life. The CINDERELLA Project pioneers an artificial intelligence (AI)-driven approach (CINDERELLA APP) for shared decision-making process, aiming to harmonise locoregional therapeutic interventions with breast cancer patients' expectations about aesthetic outcomes. The CINDERELLA clinical trial aims to establish a new standard in patient-centred care by bridging the gap between clinical treatment options and patient expectations through innovative technology. The trial focuses on evaluating the effectiveness of the CINDERELLA APP in improving patient satisfaction regarding locoregional treatment aesthetic outcomes, aligning patient expectations with real-world results, and assessing its impact on overall quality of life and psychological well-being. Methods: Trial design and statistical methods: This international multicentric interventional randomised controlled open-label clinical trial will recruit and randomise patients into two groups: one receiving standard treatment information and the other using the AI-powered CINDERELLA APP. The primary objective is to assess the levels of agreement among patients' expectations regarding the aesthetic outcome before and 12 months after locoregional treatment. The trial will also evaluate the aesthetic outcome level of agreement between the AI evaluation tool and self-evaluation. The impact of the intervention on aligning expectations with outcomes will be evaluated using the Wilcoxon signed-rank test. The improvement in classification of aesthetic results post-intervention will be measured by calculating the Weighted Cohen's kappa. Outcomes across different groups will be compared using statistical tests and bootstrap methods. CANKADO functions as the base system, allowing doctors to supervise APP content for patients and handle data gathering, while upholding principles of privacy, data security, and ethical AI practices. Intervention planned: Using the CINDERELLA APP, the patient will have access to supervised medical information approved by breast cancer experts, and the AI system will match patient's information to pictures showing the potential aesthetic outcome (spectrum of good-poor) according to different locoregional approach. Major eligibility criteria: Non-metastatic breast cancer patients eligible for either breast-conserving surgery or mastectomy with immediate reconstruction. Current enrollment: Recruitment is currently open at six study sites. The recruitment started on 8 August 2023, aiming to enroll at least 515 patients/arm. As of January 26, 2024, clinical study sites have successfully randomised 177 patients. Clinical trial information: NCT05196269 .

Abstract