Two-center study of 127 nuclear cataract cases: AS-OCT image features correlate with LOCS III grading, with automated grading accuracy of 0.81 and 0.87

AS-OCT pixel-based features significantly correlate with LOCS III grading of nuclear cataract, with mean, variance, root mean square, interquartile range, and percentiles reaching P < .01. Nuclear cataract grading has relied on subjective standards such as LOCS III; this work links automatically extractable pixel statistics from AS-OCT images to that grading standard quantitatively. Based on AS-OCT images from 127 individuals with different NC severity across two clinical centers, all graded under the LOCS III standard, with statistical association analysis reporting P < .01.

Features differ in the direction of their consistency with grading: variance, standard deviation, and median showed high consistency, while kurtosis and skewness were negatively correlated. Beyond listing which features correlate, the study distinguishes the direction of each feature's consistency with grading, informing later selection of quantitative indicators. Derived from AS-OCT pixel feature analysis of the same 127 cases, described in the text in terms of high consistency and negative correlation.

Automated machine learning models extracted nuclear region annotation and, on that basis, classified nuclear cataract into three severity stages, reaching 0.81 and 0.87 accuracy at the two centers. Combining automated nuclear region annotation with feature quantification yields a grading pipeline evaluated across centers, rather than single-center feature description alone. Validated separately at two independent clinical centers: SZRM accuracy 0.81 with F1 0.82; Thai center accuracy 0.87 with F1 0.83.

The authors conclude that automated AS-OCT image features have strong consistency in lens opacity grading and show potential for supportive diagnosis and surgical planning in NC. Extends image feature analysis from the grading task toward clinical supportive diagnosis and surgical planning applications. Conclusions rest on the two-center feature correlations and grading model performance under a clinical validation design, with the text stating that potentials are also shown.

The study targets patients with nuclear cataract (NC) and applies to clinical settings where AS-OCT examination is available and LOCS III grading is used, such as severity assessment and preoperative planning discussions in cataract clinics. It enables follow-up work to build on the automated nuclear region annotation and feature quantification pipeline validated at two centers, exploring pixel statistical features for grading support; for clinicians, this means AS-OCT images can offer computable opacity indicators beyond morphological observation. The conclusions are explicitly scoped to nuclear cataract and three-stage severity classification.

A careful reader would still watch how stable the pixel feature-grading correlations are across different devices and populations; whether the negative correlation direction of kurtosis and skewness holds in larger samples; and whether the difference between the two centers' accuracies (0.81 and 0.87) relates to population or acquisition conditions. In addition, this is a fast-parse text without figures or tables and without finer grading distribution information, so the sample composition and feature distributions at each severity level cannot be further described; these remain open questions to be filled in by the original figures and tables.