Spectropathology-corroborated multimodal quantitative imaging biomarkers for neuroretinal degeneration in diabetic retinopathy

Figures & data

Figure 1 Retinal OCT images showing six layers and their thicknesses highlighted in different colors.

Notes: RNFL, red; GCL, blue; INL, orange; OPL, pink; ONL + IS, sky blue; RPE, green. Scale bar 200 μm.
Abbreviations: GCL, ganglion cell layer; INL, inner nuclear layer; IS, inner segment layer; OCT, optical coherence tomography; ONL, outer nuclear layer; OPL, outer plexiform layer; RNFL, retinal nerve fiber layer; RPE, retinal pigment epithelium.

Table 1 Features considered during OCT image analysis

Feature type	Feature considered	References
Intensity	1. Mean gray	^Citation45
	2. Median gray
	3. Standard deviation gray
	4. Coefficient of variance gray
	5. Entropy gray
	6. Kurtosis gray
	7. Skewness gray
	8. Variance gray
Texture	9. Contrast of GLCM	^Citation17, ^Citation45
	10. Correlation of GLCM
	11. Entropy of GLCM
	12. Energy of GLCM
	13. Homogeneity of GLCM
	14. Cluster prominence
	15. Cluster shade
	16. Maximum probability
	17. Information measures of correlation
	18. Sum of variance
	19. Sum of entropy
	20. Difference entropy
	21. LBP, standard deviation
	22. LBP, mean

Abbreviations: GLCM, gray-level co-occurrence matrix; LBP, local binary pattern.

Figure 2 LDA score plot of OCT intensity and textural features using 20 principal components after PCA–LDA, with the confidence ellipse representing confidence interval at 95%.

Abbreviations: DM, diabetes mellitus; DR, diabetic retinopathy; LDA, linear discriminant analysis; LD1, linear discriminant 1; NOM, normal; OCT, optical coherence tomography; PCA, principal component analysis.

Figure 3 (A) Schematic diagram showing the first bifurcation point of the superior and inferior arterioles originating from the optic disk, considered during tortuosity, radius of curvature, and curvature measurement. (B) Clinically, two lines crossing vertically through the center of the optic disk, one from 12 o’clock to 6 o’clock and the other from 3 o’clock to 9 o’clock, were taken to define the four quadrants clockwise, namely, superior nasal (1), superior temporal (2), inferior temporal (3) and inferior nasal (4).

Notes: (A) Reproduced from Mazumder AG, Sharma UR, Aishwaryaprajna, Nawn D, Chakraborty D, Chatterjee J. Extracting arteriolar geometric attributes (tortuosity index, curvature, bifurcation angles) in normal and diabetic colour fundus images-a preliminary report. Current Indian Eye Research. 2015; Dec 2015:86–87.⁵⁰

Figure 4 Mean FTIR spectra of the whole region (400–4,000 cm⁻¹).

Notes: (A) Mean spectra of the fingerprint region (900–1,800 cm⁻¹) after RBBC; (B) mean spectra of the region between 400 and 1,200 cm^–1 after RBBC, maximum vector normalization, followed by Savitzky–Golay differentiation of the first derivative spectra of NOM, DM, and DR; (C) LDA score plot of preprocessed spectra after mean centering and (D) PCA–LDA with confidence ellipse representing confidence interval at 95%. Mean spectra of the whole region for all conditions: (E) NOM, (F) DM, and (G) DR.
Abbreviations: au, arbitrary unit; DM, diabetes mellitus; DR, diabetic retinopathy; FTIR, Fourier transform infrared; LDA, linear discriminant analysis; LD1, linear discriminant 1; LD2, linear discriminant 2; NOM, normal; PCA, principal component analysis; RBBC, rubber band-like baseline correction.

Figure 5 Confusion matrix of multiclass disease classification using intensity and textural features extracted from retinal OCT images by cubic SVM at 10-fold cross-validation.

Abbreviations: DM, diabetes mellitus; DR, diabetic retinopathy; FNR, false negative ratio; NOM, normal; OCT, optical coherence tomography; SVM, support vector machine; TPR, true positive ratio.

Table 2 Irrelevant features identified from OCT images after sequential feature reduction technique using Weka during QIB selection

Disease classified	Feature not used (feature number used in Table 1)
NOM vs DR	1,2,3,4,5,6,7,8,12,14,15,17,18,19,20,21,22
DM vs DR	3,4,5,6,7,9,10,17,19,20,21,22
NOM vs DM vs DR	1,2,3,4,5,6,7,8,9,10,11,16,18,19,20,21,22

Abbreviations: DM, diabetes mellitus; DR, diabetic retinopathy; NOM, normal; OCT, optical coherence tomography; QIB, quantitative imaging biomarkers.

Figure 6 Interconnected box plots showing intraretinal layer thickness measurements (in micrometers) in the pericentral area of the macula in patients with type 2 diabetes with no DR (DM) or DR compared to normal healthy individuals (NOM).

Note: The error bars indicate span of the data and also indicate the variability of data.
Abbreviations: DR, diabetic retinopathy; DM, diabetes mellitus; GCL, ganglion cell layer; INL, inner nuclear layer; IS, inner segment layer; NOM, normal healthy condition; ONL, outer nuclear layer; OPL, outer plexiform layer; RNFL, retinal nerve fiber layer; RPE, retinal pigment epithelium.

Table 3 Mean intraretinal layer thickness measurements (in micrometers) in the pericentral area of the macula in patients with type 2 diabetes with no DR and in patients with DR compared to healthy individuals

Parameters	NOM	DM	DR
	(n=15)	(n=15)	(n=15)
1) RNFL	23.4±3.2	22.7±2.1	21.1±3.2
2) GCL	48.8±7.2	45.2±6.1	42.5±7.2
3) ONL + IS	93.9±8.4	89.3±9.1	88.7±9.1
4) INL	38.2±3.1	38.9±3.4	39.2±3.1
5) OPL	27.8±3.0	28.1±2.7	27.9±3.1
6) RPE	39.5±1.7	39.2±1.9	38.8±1.6

Note: Values are the mean value ± SD for all subjects in each group.

Abbreviations: DM, diabetes mellitus; DR, diabetic retinopathy; GCL, ganglion cell layer; INL, inner nuclear layer; IS, inner segment layer; NOM, normal; NL, outer nuclear layer; ONL, outer nuclear layer; OPL, outer plexiform layer; RNFL, retinal nerve fiber layer; RPE, retinal pigment epithelium.

Figure 7 Box plots showing differences among the groups.

Notes: (A) Retinal arteriolar tortuosity indexes; (B) radii of curvature; and (C) curvatures among NOM, DM, and DR individuals. The error bars indicate span of the data and also indicate the variability of data.
Abbreviations: NOM, normal healthy condition; DM, diabetes mellitus; DR, diabetic retinopathy.

Table 4 Classification performance of alternates of SVM based on texture and intensity attributes extracted from OCT images

Classification condition	Classifier used	Sensitivity (%)	Specificity (%)	Accuracy (%)
NOM vs DR	Linear	70.0	82.8	79.5
	Quadratic	90.9	92.9	92.3
	Cubic	91.7	96.3	94.9
DM vs DR	Linear	71.4	73.7	72.7
	Quadratic	76.5	87.5	81.8
	Cubic	93.3	94.4	93.9
NOM vs DM vs DR	Linear	71.4	80.6	78.3
	Quadratic	78.6	90.0	83.3
	Cubic	92.9	93.1	91.7

Note: Bold values indicate best classification accuracies found during feature selection.

Abbreviations: DM, diabetes mellitus; DR, diabetic retinopathy; NOM, normal; OCT, optical coherence tomography; SVM, support vector machine.

Figure 8 Schematic diagram showing (A) abnormal polyol metabolism in progressive retinal neurodegeneration and development of diabetic retinopathy; (B) impaired retinol metabolism, and (C) role of UDP-Glc-NAc in progressive neuroretinal degeneration.

Abbreviations: O-Glc-NAc, O-linked N-acetyl glucosamine; RBP, retinol-binding protein; Raldh3, a retinaldehyde dehydrogenase that generates retinoic acid; TCA, tricarboxylic acid; UDP-Glc-NAc, uridine 5′-diphosphate-N-acetyl glucosamine; RXR, retinoid X receptor; RA, retinoic acid; atRA, all-trans retinoic acid; 9cRA, 9-cis-retinoic acid; NAD+, nicotinamide adenine dinucleotide+; NADH, nicotinamide adenine dinucleotide (reduced); FADH₂, flavin adenine dinucleotide (reduced).

Figure 9 Stacked ¹H-NMRspectra.

Notes: Spectra of (A) Normal, (B) DM-affected, and (C) DR-affected individuals.1: UDP-Glc-NAc; 2: ribitol; 3: glycerophosphocholine; and 4: fructose-6-phosphate.
Abbreviations: DM, diabetes mellitus; DR, diabetic retinopathy; UDP-Glc-NAc, uridine 5′-diphosphate-N-acetyl glucosamine.

Figure 10 ROC curve generated from the ¹H-NMR spectral data to identify serum metabolomic biomarkers for neuroretinal degeneration.

Notes: (A) Fructose-6-phosphate, (B) glycerophosphocholine, (C) ribitol, (D) UDP-Glc-NAc of DM and DR, showing the fold change of metabolites, with sensitivity and specificity of classification model in discriminating between DM and DR.
Abbreviations: AUC, area under the curve; DM, diabetes mellitus; DR, diabetic retinopathy; NMR, nuclear magnetic resonance; ROC, receiver operating characteristic; UDP-Glc-NAc, uridine 5′-diphosphate-N-acetyl glucosamine.

Table 5 Chemical shifts of major metabolites identified in ¹H-NMR spectra through comparison with HMDB

Metabolites	^1H-NMR chemical shifts (ppm)		Altered status in (nonproliferative) diabetic retinopathy (early condition)	Spectra type
	Found	HMDB
Ribitol	3.981	3.8	Upregulated	Multiplet
	3.682	3.68	Unchanged	Singlet
	3.895	3.83	Upregulated	Multiplet
	3.707	3.70	Upregulated	Singlet
Fructose-6-phosphate	3.945	3.94	Upregulated	Singlet
	3.54	3.54	Unchanged	Singlet
UDP-Glc-NAc	3.552	3.553	Upregulated	Singlet
	3.893	3.867	Unchanged	Multiplets
	3.922	3.916	Upregulated	Multiplets
Glycerophosphocholine	3.642	3.645	Downregulated	Singlet
	3.591	3.598	Unchanged	Singlet
	3.58	3.572	Downregulated	Singlet

Notes: As extension of our initial report, present study performed on larger sample size along with generation of ROC curve for ¹H-NMR spectral data and explored plausible involvement of DR related metabolic pathways. Reproduced from Mazumder AG, Ghosh S, Bag S, Bera S, Ghosh S, Mukherjee A, Chatterjee J. ¹H-NMR based serum metabolomics signatures imperative in retinal neurodegeneration and development of Diabetic Retinopathy. Int J Med Res Rev. 2016;4(6):976–981.⁴⁹

Abbreviations: HMDB, Human Metabolome Database; NMR, nuclear magnetic resonance; UDP-Glc-NAc, uridine diphosphate-N-acetyl glucosamine.

Figure 11 Representation of multimodal characterization of retinal layers with plausible spectropathological information validating the complementarity of approaches.

Notes: Preliminary data of these experiments have been accepted as conference proceeding/abstract and presented at the Mechanisms of Neurodegeneration conference at EMBL (European Molecular Biology Laboratory), Heidelberg, Germany (14th–17th June, 2017) and at the Cell Death conference at Cold Spring Harbor Laboratory, New York, USA (15th–19th August, 2017).
Abbreviations: AF, automated fragmentation; CCD, charge-coupled device; DM, diabetes mellitus; DR, diabetic retinopathy; FA, fluorescein angiography; FTIR, Fourier transform infrared; NMR, nuclear magnetic resonance; OCT, optical coherence tomography; QIB, quantitative imaging biomarker; RF, radio frequency; SLD, superluminescent diode.

RuanH-BNieYYangXRegulation of protein degradation by O-GlcNAcylation: crosstalk with ubiquitinationMol Cell Proteomics201312123489349723824911

 PubMed Web of Science ®Google Scholar

KurJNewmanEAChan-LingTCellular and physiological mechanisms underlying blood flow regulation in the retina choroid in health diseaseProg Retin Eye Res201231537740622580107

 PubMed Web of Science ®Google Scholar

MazumderAGSharmaURAishwaryaprajnaNawnDChakrabortyDChatterjeeJExtracting arteriolar geometric attributes (tortuosity index, curvature, bifurcation angles) in normal and diabetic colour fundus images-a preliminary reportCurrent Indian Eye Research20151220158687

 Google Scholar

MazumderAGGhoshSBagSBeraSGhoshSMukherjeeAChatterjeeJ1H-NMR based serum metabolomics signatures imperative in retinal neurodegeneration and development of Diabetic RetinopathyInt J Med Res Rev201646976981

 Google Scholar