Reduced macula microvascular densities may be an early indicator for diabetic peripheral neuropathy

Abstract

Purpose: To assess the alteration in the macular microvascular in type 2 diabetic patients with peripheral neuropathy (DPN) and without peripheral neuropathy (NDPN) by optical coherence tomography angiography (OCTA) and explore the correlation between retinal microvascular abnormalities and DPN disease. Methods: Twenty-seven healthy controls (42 eyes), 36 NDPN patients (62 eyes), and 27 DPN patients (40 eyes) were included. OCTA was used to image the macula in the superficial vascular complex (SVC) and deep vascular complex (DVC). In addition, a state-of-the-art deep learning method was employed to quantify the microvasculature of the two capillary plexuses in all participants using vascular length density (VLD). Results: Compared with the healthy control group, the average VLD values of patients with DPN in SVC (p = 0.010) and DVC (p = 0.011) were significantly lower. Compared with NDPN, DPN patients showed significantly reduced VLD values in the SVC (p = 0.006) and DVC (p = 0.001). Also, DPN patients showed lower VLD values (p < 0.05) in the nasal, superior, temporal and inferior sectors of the inner ring of the SVC when compared with controls; VLD values in NDPN patients were lower in the nasal section of the inner ring of SVC (p < 0.05) compared with healthy controls. VLD values in the DVC (AUC = 0.736, p < 0.001) of the DPN group showed a higher ability to discriminate microvascular damage when compared with NDPN. Conclusion: OCTA based on deep learning could be potentially used in clinical practice as a new indicator in the early diagnosis of DM with and without DPN.

Keywords: diabetic peripheral neuropathy; diabetic retinopathy; microvasculature; optical coherence tomography angiography; vascular length density.

FIGURE 1

OCTA image analysis of macular fovea and artificial intelligence algorithm for layered analysis of images. (A) Using OCTA to scan the macula area of the subject within the range of 3 × 3 mm², and the images of capillaries around the fovea are obtained. Then, using deep learning software, FAZ was used to fit a circle with radius r, after which three concentric circles with radius r, 1.5 r, and 2 r were drawn. With further partitioning, the retina was divided into inner and outer rings: nasal area, superior area, temporal area, and inferior area, and there were 8 areas in total. (B) The retina was segmented into SVC (from the internal limiting membrane to 10 μm above the inner plexiform layer) and DVC (from 10 μm above the inner plexiform layer to the 10 μm below the outer plexiform layer) images with different depths. The images (C–K) show retinal microvascular in a different layer of HC, NDPN, and DPN groups. (C–E) The images show the full-thickness retinal microvasculature produced by OCTA in HC, NDPN, and DPN groups. (F–H) Capillaries in the SVC by deep learning software. (I–K) Capillaries in the DVC by deep learning software.

FIGURE 2

Comparison of VLD in different sectors between HC, NDPN, and DPN groups. (A) Describe the VLD of the SVC in four inner quadrant sectors (nasal inner, superior inner, temporal inner, and inferior inner). (B) VLD of DVC in four inner sectors. (C) VLD of SVC in four outer quadrant sectors (nasal outer, superior outer, temporal outer, and inferior outer). (D) VLD of DVC in four outer quadrant sectors. VLD is defined as the ratio of the total number of pixels on microvascular centerlines to the area of measurement.

FIGURE 3

ROC curve analysis of VLD in the SVC and DVC. (A) The diagnostic efficiency of VLD value of NDPN group in different levels. (B) The diagnostic efficiency VLD value of the DPN group. Greenline: The ROC of DVC. Blueline: The ROC of SVC.