Scholarlink

Background/aims <p>Early recognition of eyelid morphological abnormalities was crucial, as untreated conditions could lead to blinding complications. An eyelid screening system that could provide both anatomical and pathological information was essential for formulating personalized treatment strategies. This study aimed to develop a clinically concerns-based framework capable of identifying common eyelid diseases requiring further intervention by evaluating individual anatomical and pathological changes. This approach would enhance individualized and efficient prevention, while supporting targeted treatment strategies.</p> Methods <p>The eyelid disorder screening system, Eyetome, was developed based on a morphological atlas and comprised four modules designed to identify 14 common eyelid disorders and pathological changes. A total of 6180 eye patches were analyzed to extract anatomical and pathological features. The performance of Eyetome was evaluated using average accuracy (aACC) and F1 score, with comparisons made against traditional models and ophthalmologists. To assess the system’s expandability, an additional test was conducted in a multimorbidity scenario.</p> Results <p>Eyetome demonstrated high performance in recognizing single diseases, achieving an aACC of 98.83% and an F1 score of 0.93. The system outperformed classic models, with an aACC of 98.83% compared to 96.72% for Desnet101 and 97.59% for Vit. Additionally, Eyetome’s aACC exceeded that of a junior ophthalmologist (JO) (97.11%) and was comparable to a senior ophthalmologist (SO) (98.69%). In the extended multimorbidity dataset, Eyetome maintained robust performance with an accuracy of 97.97%, surpassing JO (95.47%) and closely matching SO (97.81%).</p> Conclusions <p>This study developed a clinical concerns-based system for screening and monitoring eyelid disorders, aimed at supporting predictive diagnosis, preventing diseases progression, and facilitating more effective, patient-centered treatment of common eyelid disorders, aligning with the principles of predictive, preventive, and personalized medicine (PPPM/3PM). The system’s interpretability, scalability, and user-friendly data acquisition design could further enhance its acceptance among both doctors and patients, facilitating the shift from reactive medicine to proactive precision medicine.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advancing predictive, preventive, and personalized medicine in eyelid diseases: a concerns-based and expandable screening system through structural dissection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年3月5日

查看全文
  • Jing Cao,
  • Kun You,
  • Peifang Xu,
  • Yiming Sun,
  • Ji Shao,
  • Yifan Zhou,
  • Huimin Li,
  • Lixia Lou,
  • Qi Miao,
  • Juan Ye

摘要

Background/aims

Early recognition of eyelid morphological abnormalities was crucial, as untreated conditions could lead to blinding complications. An eyelid screening system that could provide both anatomical and pathological information was essential for formulating personalized treatment strategies. This study aimed to develop a clinically concerns-based framework capable of identifying common eyelid diseases requiring further intervention by evaluating individual anatomical and pathological changes. This approach would enhance individualized and efficient prevention, while supporting targeted treatment strategies.

Methods

The eyelid disorder screening system, Eyetome, was developed based on a morphological atlas and comprised four modules designed to identify 14 common eyelid disorders and pathological changes. A total of 6180 eye patches were analyzed to extract anatomical and pathological features. The performance of Eyetome was evaluated using average accuracy (aACC) and F1 score, with comparisons made against traditional models and ophthalmologists. To assess the system’s expandability, an additional test was conducted in a multimorbidity scenario.

Results

Eyetome demonstrated high performance in recognizing single diseases, achieving an aACC of 98.83% and an F1 score of 0.93. The system outperformed classic models, with an aACC of 98.83% compared to 96.72% for Desnet101 and 97.59% for Vit. Additionally, Eyetome’s aACC exceeded that of a junior ophthalmologist (JO) (97.11%) and was comparable to a senior ophthalmologist (SO) (98.69%). In the extended multimorbidity dataset, Eyetome maintained robust performance with an accuracy of 97.97%, surpassing JO (95.47%) and closely matching SO (97.81%).

Conclusions

This study developed a clinical concerns-based system for screening and monitoring eyelid disorders, aimed at supporting predictive diagnosis, preventing diseases progression, and facilitating more effective, patient-centered treatment of common eyelid disorders, aligning with the principles of predictive, preventive, and personalized medicine (PPPM/3PM). The system’s interpretability, scalability, and user-friendly data acquisition design could further enhance its acceptance among both doctors and patients, facilitating the shift from reactive medicine to proactive precision medicine.