Quantifying frequency content in cross-sectional retinal scans of diabetics vs. controls

Purpose: To examine subtle differences in the structure of diabetic vs. control retinas. Methods: Spectral-domain optical coherence tomography (SD-OCT) images were compared for the retinas of 33 diabetic subjects who did not have clinical evidence of diabetic macular edema and age-matched controls, with central macular thicknesses of 275 and 276 microns, respectively. Cross-sectional retinal images through the fovea, called B-scans, were analyzed for spatial frequency content. The B-scans were processed to remove and smooth the portions of the retinal image not within regions of interest in the retina. The remaining retinal images were then quantified using a Fast Fourier Transform (FFT) approach that provided amplitude as a function of spatial frequency. Results: The FFT analysis showed that diabetic retinas had spatial frequency content with significantly higher power compared to control retinas particularly for a deeper fundus layer at mid-range spatial frequencies, ranging from p=0.0030 to 0.0497 at 16.8 to 18.2 microns/cycle. There was lower power at higher spatial frequencies, ranging from p=0.0296 and 0.0482 at 27.4 and 29.0 microns/cycle. The range of mid-range frequencies corresponds to the sizes of small blood vessel abnormalities and hard exudates. Retinal thickness did not differ between the two groups. Conclusions: Diabetic retinas, although not thicker than controls, had subtle but quantifiable pattern changes in SD-OCT images particularly in deeper fundus layers. The size range and distribution of this pattern in diabetic eyes were consistent with small blood vessel abnormalities and leakage of lipid and fluid. Feature-based biomarkers may augment retinal thickness criteria for management of diabetic eye complications, and may detect early changes.研究目的：探究糖尿病视网膜与健康对照组视网膜在结构上的细微差异。 研究方法：选取33名无临床糖尿病黄斑水肿证据的糖尿病受试者，以及年龄匹配的健康对照组受试者，两组受试者的中心黄斑厚度分别为275微米与276微米。对其视网膜开展光谱域光学相干断层扫描（Spectral-domain optical coherence tomography, SD-OCT）成像并进行对比分析。对通过黄斑中心凹的视网膜断层图像（即B扫描（B-scans））进行空间频率成分分析：首先对B扫描图像进行预处理，去除并平滑视网膜非感兴趣区域的图像部分；随后采用快速傅里叶变换（Fast Fourier Transform, FFT）方法对剩余视网膜图像进行量化分析，得到以空间频率为自变量的振幅数据。 研究结果：快速傅里叶变换（FFT）分析显示，糖尿病视网膜的空间频率成分功率显著高于对照组，尤其在中等空间频率范围的深层眼底层表现突出，在16.8至18.2微米/周期的区间内，p值范围为0.0030至0.0497。在较高空间频率区间，糖尿病视网膜的功率则更低，在27.4与29.0微米/周期处的p值分别为0.0296与0.0482。该中等频率区间的尺寸范围与小型血管异常和硬性渗出物的尺寸相符。两组受试者的视网膜厚度无显著差异。 研究结论：尽管糖尿病视网膜的厚度并未高于健康对照组，但其光谱域光学相干断层扫描（SD-OCT）图像中存在细微但可量化的模式变化，尤以深层眼底层为显著。糖尿病患者眼部的这种模式的尺寸范围与分布特征，与小型血管异常以及脂质和液体渗漏的情况相一致。基于特征的生物标志物可作为视网膜厚度评估标准的补充，用于糖尿病眼部并发症的临床管理，并有望实现早期病变检测。