上皮泽尼克指数和人工智能可区分有瓣和无瓣屈光手术中上皮重塑。

Epithelium Zernike Indices and Artificial Intelligence Can Differentiate Epithelial Remodeling Between Flap and Flapless Refractive Procedures.

出版信息

J Refract Surg. 2020 Feb 1;36(2):97-103. doi: 10.3928/1081597X-20200103-01.

DOI:10.3928/1081597X-20200103-01

PMID:32032430

Abstract

PURPOSE

To evaluate epithelial Zernike indices as a differentiator of epithelial remodeling after different refractive procedures.

METHODS

Optical coherence tomography (OCT) images of 22 laser in situ keratomileusis, 22 small incision lenticule extraction, 15 photorefractive keratectomy (PRK), and 17 transepithelial PRK eyes were evaluated retrospectively before and after surgery. A custom algorithm was used to calculate the epithelial Zernike indices from the three-dimensional distribution of epithelial thickness distribution. The epithelial Zernike indices were also compared with the local measurements of epithelial thickness, used conventionally from the current clinical OCT. A decision tree classifier was built, one in which flap/cap and surface procedures were classified (2G) and another in which all surgical groups were classified separately (4G).

RESULTS

Local measurements of thicknesses changed significantly after all surgeries (P < .05), but these changes were similar in magnitude between the surgical platforms (P > .05). The surgeries not only changed the epithelial Zernike indices (P < .05), but also resulted in differential changes in epithelial thickness distribution based on the type of surgery (P < .05). In the 2G analyses with local measurements of epithelial thickness, the area under the curve, sensitivity, and specificity were 0.57 ± 0.07, 42.11%, and 57.89%, respectively. Further, the accuracy was limited to less than 60%. In the 2G analyses with epithelial Zernike indices, the area under the curve, sensitivity, and specificity were 0.79 ± 0.05, 86.4%, and 71.9%, respectively. Here, the accuracy was limited between 70% and 80%. Similar trends were observed with 4G analyses.

CONCLUSIONS

The epithelial Zernike indices were significantly better in identifying surgery-specific three-dimensional remodeling of the thickness compared to local measurements of epithelial thickness. Further, the changes in Zernike indices were independent of the magnitude of refractive error but not the type of surgery. [J Refract Surg. 2020;36(2):97-103.].

摘要

目的

评估上皮泽尼克指数是否可作为区分不同屈光手术后上皮重塑的指标。

方法

回顾性分析 22 例激光原位角膜磨镶术、22 例小切口微透镜提取术、15 例准分子光角膜切削术和 17 例经上皮准分子光角膜切削术的患者术前和术后的光学相干断层扫描（OCT）图像。使用定制算法，从上皮厚度分布的三维分布中计算上皮泽尼克指数。还将上皮泽尼克指数与传统临床 OCT 中常用的上皮厚度的局部测量值进行比较。建立了一个决策树分类器，一个用于分类瓣/帽和表面手术（2G），另一个用于分类所有手术组（4G）。

结果

所有手术术后上皮厚度的局部测量值均发生显著变化（P <.05），但不同手术平台之间的变化幅度相似（P >.05）。这些手术不仅改变了上皮泽尼克指数（P <.05），而且还根据手术类型导致上皮厚度分布的差异变化（P <.05）。在使用上皮厚度局部测量值的 2G 分析中，曲线下面积、敏感性和特异性分别为 0.57 ± 0.07、42.11%和 57.89%。此外，准确性限制在 60%以下。在使用上皮泽尼克指数的 2G 分析中，曲线下面积、敏感性和特异性分别为 0.79 ± 0.05、86.4%和 71.9%。在这里，准确性限制在 70%到 80%之间。4G 分析也观察到类似的趋势。