Focal Loss Analysis of Nerve Fiber Layer Reflectance for Glaucoma Diagnosis

Glaucoma Detection through a Novel Hyperspectral Imaging Band Selection and Vision Transformer Integration

Conventional diagnostic methods for glaucoma primarily rely on non-dynamic fundus images and often analyze features such as the optic cup-to-disc ratio and abnormalities in specific retinal locations like the macula and fovea. However, hyperspectral imaging techniques focus on detecting alterations in oxygen saturation within retinal vessels, offering a potentially more comprehensive approach to diagnosis. This study explores the diagnostic potential of hyperspectral imaging for glaucoma by introducing a novel hyperspectral imaging conversion technique. Digital fundus images are transformed into hyperspectral representations, allowing for a detailed analysis of spectral variations. Spectral regions exhibiting differences are identified through spectral analysis, and images are reconstructed from these specific regions. The Vision Transformer (ViT) algorithm is then employed for classification and comparison across selected spectral bands. Fundus images are used to identify differences in lesions, utilizing a dataset of 1291 images. This study evaluates the classification performance of models using various spectral bands, revealing that the 610-780 nm band outperforms others with an accuracy, precision, recall, F1-score, and AUC-ROC all approximately at 0.9007, indicating its superior effectiveness for the task. The RGB model also shows strong performance, while other bands exhibit lower recall and overall metrics. This research highlights the disparities between machine learning algorithms and traditional clinical approaches in fundus image analysis. The findings suggest that hyperspectral imaging, coupled with advanced computational techniques such as the ViT algorithm, could significantly enhance glaucoma diagnosis. This understanding offers insights into the potential transformation of glaucoma diagnostics through the integration of hyperspectral imaging and innovative computational methodologies.

A Hybrid Deep Learning Classification of Perimetric Glaucoma Using Peripapillary Nerve Fiber Layer Reflectance and Other OCT Parameters from Three Anatomy Regions

Précis

A hybrid deep-learning model combines NFL reflectance and other OCT parameters to improve glaucoma diagnosis.

Objective

To investigate if a deep learning model could be used combine nerve fiber layer (NFL) reflectance and other OCT parameters for glaucoma diagnosis.

Patients and Methods

This is a prospective observational study where of 106 normal subjects and 164 perimetric glaucoma (PG) patients. Peripapillary NFL reflectance map, NFL thickness map, optic head analysis of disc, and macular ganglion cell complex thickness were obtained using spectral domain OCT. A hybrid deep learning model combined a fully connected network (FCN) and a convolution neural network (CNN) to develop to combine those OCT maps and parameters to distinguish normal and PG eyes. Two deep learning models were compared based on whether the NFL reflectance map was used as part of the input or not.

Results

The hybrid deep learning model with reflectance achieved 0.909 sensitivity at 99% specificity and 0.926 at 95%. The overall accuracy was 0.948 with 0.893 sensitivity and 1.000 specificity, and the AROC was 0.979, which is significantly better than the logistic regression models (p < 0.001). The second best model is the hybrid deep learning model w/o reflectance, which also had significantly higher AROC than logistic regression models (p < 0.001). Logistic regression with reflectance model had slightly higher AROC or sensitivity than the other logistic regression model without reflectance (p = 0.024).

Conclusions

Hybrid deep learning model significantly improved the diagnostic accuracy, without or without NFL reflectance. Hybrid deep learning model, combining reflectance/NFL thickness/GCC thickness/ONH parameter, may be a practical model for glaucoma screen purposes.

Predicting perimetric defects from en face maps of retinal nerve fibre layer reflectance

PURPOSE

To develop criteria to predict visual hemifields with deep perimetric defects based on retinal nerve fibre layer (RNFL) reflectance, in a transparent process whose components can be assessed by independent laboratories analysing data from their own small groups.

METHODS

The analysis was carried out in four stages, using three independent groups of patients-30, 33 and 62 participants-with glaucoma and age-similar controls. The first stage used Group 1 to develop a criterion for RNFL reflectance images at 24, 36 or 48 μm below the inner limiting membrane (ILM). The second stage evaluated the criterion using Group 2. The third stage developed a second criterion to improve performance for Groups 1 and 2 combined. The fourth stage evaluated the second criterion with Group 3. Confidence intervals for sensitivity and specificity were then computed by combining results from all three groups.

RESULTS

The first criterion identified all hemifields with deep defects and no hemifields from controls, using a within-eye reference for healthy RNFL. For Group 2, specificity remained high but sensitivity was reduced. The second criterion improved sensitivity by using location-specific reference values. For Group 3, sensitivity remained high but reduced specificity was found. Confidence intervals showed substantial overlap for the two criteria.

CONCLUSIONS

We developed two criteria to identify patients with deep perimetric defects with high specificity and sensitivity. Several improvements are warranted: automated identification of the fovea-disc angle and optic disc locations, evaluation of normal variation in patterns of RNFL thickness, improved segmentation of ILM and major vasculature, reduction of within-eye variability in RNFL reflectance of healthy eyes, assessment of effects of image quality, assessment of effects of comorbidity and effectiveness of other devices.

Structure-function assessment in glaucoma based on perimetric sensitivity and en face optical coherence tomography images of retinal nerve fiber bundles

Many studies have assessed structure-function relations in glaucoma, but most without topographical comparison across the central 30°. We present a method for assessing structure-function relations with en face images of retinal nerve fiber layer (RNFL) bundles allowing topographical comparison across much of this retinal area. Forty-four patients with glaucoma (median age 61 years) were recruited and tested with Optical Coherence Tomography (OCT) and perimetry. Six rectangular volume scans were gathered, and then montaged to provide en face views of the RNFL bundles. We calculated the proportion of locations showing a perimetric defect that also showed an en face RNFL defect; and the proportion of locations falling on an RNFL defect that also showed a perimetric defect. A perimetric defect for a location was defined as a total deviation (TD) value equal to or deeper than -4 dB. We found that the median (IQR) number of locations with abnormal RNFL bundle reflectance that also had abnormal TD was 78% (60%) and for locations with abnormal TD that also had abnormal RNFL bundle reflectance was 75% (44%). We demonstrated a potential approach for structure-function assessment in glaucoma by presenting a topographic reflectance map, confirming results of previous studies and including larger retinal regions.

Concordance of Objectively Detected Retinal Nerve Fiber Bundle Defects in En Face OCT Images with Conventional Structural and Functional Changes in Glaucoma

PURPOSE

To assess how objectively detected defects in retinal nerve fiber bundle (RNFB) reflectance on en face OCT images relate to circumpapillary retinal nerve fiber layer thickness (cpRNFLT) and visual field defects.

DESIGN

Cross-sectional study.

PARTICIPANTS

Sixteen participants with early glaucoma and 29 age-matched healthy controls, of whom 22 had usable en face images for the establishment of normative levels of RNFB reflectance.

METHODS

All the participants underwent cpRNFLT scans, visual field examination, and wide-field OCT. En face reflectivity was assessed objectively using the Summary of Multiple Anatomically Adjusted Slabs method. En face defects were deemed concordant with cpRNFLT when they had at least 1 cpRNFLT point with P < 0.01, within ± 15° of the predicted insertion on the optic disc. Visual fields were examined using custom suprathreshold perimetry and SITA Standard 24-2. For each visual field location, the corresponding reflectance was deemed abnormal if any en face superpixel within ± 1° was abnormal. The overall, positive, and negative agreements were measured in each participant.

MAIN OUTCOME MEASURES

Proportion of concordant defects between en face reflectance analysis and cpRNFLT (%) as well as overall, positive, and negative agreements between en face reflectance analysis and visual field results.

RESULTS

Most en face abnormalities had concordant cpRNFLT defects in the mapped sector (median proportion concordant, 0.85; interquartile range, 0.74-0.95). In eyes with glaucoma, a median of 8.1% (range, 2.4%-23.7%) and 14.9% (range, 3.5%-29.1%) locations showed corresponding en face and visual field defects using 24-2 and custom perimetry, respectively. Both the perimetric strategies had moderate-to-good raw agreement with en face analysis (0.66-0.68), with stronger agreement on normal findings than on defects (0.77-0.78 and 0.4-0.44).

CONCLUSIONS

Objectively extracted reflectance defects showed strong concordance with conventional cpRNFLT damage and good agreement with perimetry, which could be enhanced by further minimization of image artifacts.

Adaptive IoU Thresholding for Improving Small Object Detection: A Proof-of-Concept Study of Hand Erosions Classification of Patients with Rheumatic Arthritis on X-ray Images

In recent years, much research evaluating the radiographic destruction of finger joints in patients with rheumatoid arthritis (RA) using deep learning models was conducted. Unfortunately, most previous models were not clinically applicable due to the small object regions as well as the close spatial relationship. In recent years, a new network structure called RetinaNets, in combination with the focal loss function, proved reliable for detecting even small objects. Therefore, the study aimed to increase the recognition performance to a clinically valuable level by proposing an innovative approach with adaptive changes in intersection over union (IoU) values during training of Retina Networks using the focal loss error function. To this end, the erosion score was determined using the Sharp van der Heijde (SvH) metric on 300 conventional radiographs from 119 patients with RA. Subsequently, a standard RetinaNet with different IoU values as well as adaptively modified IoU values were trained and compared in terms of accuracy, mean average accuracy (mAP), and IoU. With the proposed approach of adaptive IoU values during training, erosion detection accuracy could be improved to 94% and an mAP of 0.81 ± 0.18. In contrast Retina networks with static IoU values achieved only an accuracy of 80% and an mAP of 0.43 ± 0.24. Thus, adaptive adjustment of IoU values during training is a simple and effective method to increase the recognition accuracy of small objects such as finger and wrist joints.

Within-eye and between-subject variability for reflectance of the retinal nerve fibre layer

PURPOSE

Reflectance of retinal nerve fibre layer (RNFL) can contribute to detecting the presence of glaucomatous damage and defining its extent. As a step towards developing a normative database for RNFL reflectance, we assessed within-eye and between-subject variability for RNFL reflectance in healthy eyes.

METHODS

Vertical 30° × 15° volume scans at the optic disc were gathered using SD-OCT (Spectralis OCT) from people free of eye disease. Scans were gathered for both eyes of 30 younger adults (mean ± SD = 27 ± 3 years) and for one eye of 30 older adults (68 ± 8 years). Reflectance was quantified for each voxel as the depth-resolved attenuation coefficient (AC). Values for AC were extracted for four slabs (0-52, 24-52, 24-36 and 36-60 μm) and at depths from 24 to 60 μm below the inner limiting membrane (ILM) in 4 μm steps.

RESULTS

Between-subject and within-eye standard deviations (SDs) for the logarithm of AC were similar; median differences were 0.02-0.03 log unit across all four slabs and depths from 24 to 48 μm. Means for the logarithm of AC were higher for younger than older eyes by ~0.1 log unit; this age effect was not due to differences in the raw reflectance of the RNFL, but rather to age-related changes in reflectance of deeper retina affecting the calculation of AC.

CONCLUSIONS

In both groups, within-eye variability in RNFL reflectance near the optic disc was similar to between-subject variability. A better understanding of within-eye variability would be useful for developing normative databases.

Literature Review on Artificial Intelligence Methods for Glaucoma Screening, Segmentation, and Classification

Artificial intelligence techniques are now being applied in different medical solutions ranging from disease screening to activity recognition and computer-aided diagnosis. The combination of computer science methods and medical knowledge facilitates and improves the accuracy of the different processes and tools. Inspired by these advances, this paper performs a literature review focused on state-of-the-art glaucoma screening, segmentation, and classification based on images of the papilla and excavation using deep learning techniques. These techniques have been shown to have high sensitivity and specificity in glaucoma screening based on papilla and excavation images. The automatic segmentation of the contours of the optic disc and the excavation then allows the identification and assessment of the glaucomatous disease's progression. As a result, we verified whether deep learning techniques may be helpful in performing accurate and low-cost measurements related to glaucoma, which may promote patient empowerment and help medical doctors better monitor patients.