JointViT: Modeling Oxygen Saturation Levels with Joint Supervision on Long-Tailed OCTA This paper proposes JointViT, a novel model based on the Vision Transformer architecture, which incorporates a joint loss function for supervision. The model is designed to predict oxygen saturation levels (SaO2) from optical coherence tomography angiography (OCTA) images, which are often imbalanced and long-tailed. To address the challenges posed by the long-tailed distribution of the OCTA dataset, the paper introduces a balancing augmentation technique during data preprocessing. This technique enhances the model's performance on the long-tailed distribution within the OCTA dataset. Comprehensive experiments on the OCTA dataset demonstrate that the proposed method significantly outperforms other state-of-the-art methods, achieving improvements of up to 12.28% in overall accuracy. This advancement lays the groundwork for the future utilization of OCTA in diagnosing sleep-related disorders. The paper also discusses the importance of SaO2 levels in health, particularly in relation to sleep-related breathing disorders. Continuous monitoring of SaO2 is time-consuming and highly variable depending on patients' conditions. OCTA has shown promising development in rapidly and effectively screening eye-related lesions, offering the potential for diagnosing sleep-related disorders. However, the long-tailed distribution of OCTA data makes prediction challenging. Instead of directly modeling concrete SaO2 values, the paper proposes predicting SaO2 categories, which is more robust and beneficial for the diagnosis of sleep-related breathing disorders. The paper also discusses related works in medical imaging recognition, long-tailed image recognition, and OCTA in AI for health. It highlights the importance of SaO2 prediction in diagnosing sleep apnea and other sleep-related breathing disorders. The paper presents a comprehensive evaluation of the proposed method, demonstrating its effectiveness in predicting SaO2 levels from OCTA images. The results show that the proposed method significantly outperforms other state-of-the-art methods, achieving improvements of up to 12.28% in overall accuracy. This advancement lays the groundwork for the future utilization of OCTA in diagnosing sleep-related disorders.JointViT: Modeling Oxygen Saturation Levels with Joint Supervision on Long-Tailed OCTA This paper proposes JointViT, a novel model based on the Vision Transformer architecture, which incorporates a joint loss function for supervision. The model is designed to predict oxygen saturation levels (SaO2) from optical coherence tomography angiography (OCTA) images, which are often imbalanced and long-tailed. To address the challenges posed by the long-tailed distribution of the OCTA dataset, the paper introduces a balancing augmentation technique during data preprocessing. This technique enhances the model's performance on the long-tailed distribution within the OCTA dataset. Comprehensive experiments on the OCTA dataset demonstrate that the proposed method significantly outperforms other state-of-the-art methods, achieving improvements of up to 12.28% in overall accuracy. This advancement lays the groundwork for the future utilization of OCTA in diagnosing sleep-related disorders. The paper also discusses the importance of SaO2 levels in health, particularly in relation to sleep-related breathing disorders. Continuous monitoring of SaO2 is time-consuming and highly variable depending on patients' conditions. OCTA has shown promising development in rapidly and effectively screening eye-related lesions, offering the potential for diagnosing sleep-related disorders. However, the long-tailed distribution of OCTA data makes prediction challenging. Instead of directly modeling concrete SaO2 values, the paper proposes predicting SaO2 categories, which is more robust and beneficial for the diagnosis of sleep-related breathing disorders. The paper also discusses related works in medical imaging recognition, long-tailed image recognition, and OCTA in AI for health. It highlights the importance of SaO2 prediction in diagnosing sleep apnea and other sleep-related breathing disorders. The paper presents a comprehensive evaluation of the proposed method, demonstrating its effectiveness in predicting SaO2 levels from OCTA images. The results show that the proposed method significantly outperforms other state-of-the-art methods, achieving improvements of up to 12.28% in overall accuracy. This advancement lays the groundwork for the future utilization of OCTA in diagnosing sleep-related disorders.