Scholarlink

Purpose <p>To conduct a systematic review of studies examining the use of artificial intelligence (AI) algorithms in predicting the progression of keratoconus (KCN).</p> Methods <p>A comprehensive review was conducted using PubMed, Google Scholar, and ScienceDirect, focusing on keywords related to AI and KCN progression. Only studies published in English that applied AI algorithms to evaluate changes in KCN were included. The QUADAS-2 tool was used to assess the risk of bias and the overall quality of the studies. A narrative qualitative synthesis approach was employed to summarize and interpret the findings.</p> Results <p>A total of ten studies involving 10,940 eyes were analyzed to evaluate the progression of KCN using various AI models, including machine learning and deep learning techniques. Most of these models demonstrated moderate to high discriminative ability, with reported area under the curve (AUC) values ranging from 0.77 to 0.93 and accuracies between 77.5% and 84.9%. Key predictive factors consistently identified across multiple studies included posterior elevation, maximum keratometry (Kmax), and younger age. Notably, eyes that progressed were significantly younger than those that remained stable. Furthermore, incorporating non-imaging clinical risk factors—such as IgE levels, eye rubbing, and deficiencies in vitamin D and B12—into the classification framework enhanced the performance of the models. However, none of the studies conducted external validation using an independent, multi-center cohort. Most relied on internal train-test splits or cross-validation for their analyses.</p> Conclusion <p>AI models show promise in identifying the progression of KCN. However, the current evidence is limited by inconsistent criteria for progression, insufficient external validation, dependence on specific devices, and a lack of comprehensive reporting on calibration and decision curve analysis. To integrate these tools into routine clinical practice, standardized definitions, multi-center prospective validation studies, and platform-independent algorithms are essential.</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

The application of artificial intelligence-based algorithms in predicting the progression of keratoconus: a systematic review

  • 期刊论文

  • 发表日期：

    2025年11月15日

查看全文
  • Hassan Hashemi,
  • Alireza Jamali,
  • Payam Nabovati,
  • Farshid Babapour Mofrad,
  • Mehdi Khabazkhoob

摘要

Purpose

To conduct a systematic review of studies examining the use of artificial intelligence (AI) algorithms in predicting the progression of keratoconus (KCN).

Methods

A comprehensive review was conducted using PubMed, Google Scholar, and ScienceDirect, focusing on keywords related to AI and KCN progression. Only studies published in English that applied AI algorithms to evaluate changes in KCN were included. The QUADAS-2 tool was used to assess the risk of bias and the overall quality of the studies. A narrative qualitative synthesis approach was employed to summarize and interpret the findings.

Results

A total of ten studies involving 10,940 eyes were analyzed to evaluate the progression of KCN using various AI models, including machine learning and deep learning techniques. Most of these models demonstrated moderate to high discriminative ability, with reported area under the curve (AUC) values ranging from 0.77 to 0.93 and accuracies between 77.5% and 84.9%. Key predictive factors consistently identified across multiple studies included posterior elevation, maximum keratometry (Kmax), and younger age. Notably, eyes that progressed were significantly younger than those that remained stable. Furthermore, incorporating non-imaging clinical risk factors—such as IgE levels, eye rubbing, and deficiencies in vitamin D and B12—into the classification framework enhanced the performance of the models. However, none of the studies conducted external validation using an independent, multi-center cohort. Most relied on internal train-test splits or cross-validation for their analyses.

Conclusion

AI models show promise in identifying the progression of KCN. However, the current evidence is limited by inconsistent criteria for progression, insufficient external validation, dependence on specific devices, and a lack of comprehensive reporting on calibration and decision curve analysis. To integrate these tools into routine clinical practice, standardized definitions, multi-center prospective validation studies, and platform-independent algorithms are essential.