RetinaLyze Logo

Abnormal GDF with few areas of low hemoglobin concentration

Guest post by Prof. Manuel Gonzalez de la Rosa: How can a very abnormal GDF, be compatible with the presence of only a few areas of low hemoglobin concentration?

Posted on
November 27, 2020
in
Ophthalmology

A colleague has asked us how avery abnormal GDF can be compatible with the presence of only a few areas of low hemoglobin concentration. This is the example he sent me:

 

Glaucoma, an ischemic neuropathy

Today we know that glaucoma is an ischemic neuropathy, that is, its origin lies in a defect in blood flow in the optic nerve. This perfusion defect produces an atrophy of the nerve fibers, which results in the presence of morphological changes, which can be observed with an OCT, and also functional defects that are studied by means of perimetry.

Experimental instruments have been developed to measure the origin of glaucoma in the nerve, i.e. perfusion, but they have not gone beyond the experimental level (colour laser echography, laser-doppler fluxometry, laser speckle, oximetry) and there was hope that OCT angiography (A-OCT) would provide such relevant information.

The RetinaLyze Glaucoma (also known as Laguna OHnE) studies perfusion in a very simple and highly efficient way, and makes an estimate of morphology based on perfusion, but measures both the origin of the problem and the consequences. A recent paper published in Br.J. Ophthalmology has shown that a state-of-the-art A-OCT does not outperform our method in terms of results.

 

How does RetinaLyze Glaucoma work?

Let us look at some aspects of the information provided by RetinaLyze Glaucoma:

GDF is an index developed as a multiple function that includes various elements of hemoglobin distribution in the nerve that have been shown to be useful in differentiating pathology from normality. Some of these elements are estimates of common indices, such as cup/disc ratios. Others are the concentrations of hemoglobin in areas that we have detected as critical, such as sectors 8 and 20 that we will see later.

But there is a factor that participates in an important way in the GDF that is not so easy to explain, because it is obtained by automatic training of convolutional neural networks (Deep Learning, DL) to which thousands of images of hemoglobin distribution and nerve environment, normal and glaucomatous have contributed. These networks have learned to interpret multiple aspects that cannot be clearly defined, because only one final classifier is obtained. Some characteristics that are certainly taken into account are the size and shape of the Hb cup, the displacement of the vessels towards the nasal side or peripapillary atrophy, but not all of them can be known for sure nor what is the "weight" that influences the final classifier. As can be seen, case by case, the results of this index are close to 1 in the case of normality and to 0 in the case of glaucoma, but they are not very progressive. They tend to be in extreme positions.

How results are visualized

RetinaLyze Glaucoma is a program that essentially studies perfusion, which defect we assume is the origin of glaucoma, but the results are shown as graphs and numerical data. The version of the program marketed for screening shows an overall assessment, based on the GDF index, providing three risk levels: within normal limits, at the extreme of what is considered normal or below what is expected in 99% of normal subjects.

But internally the program analyzes topographically the entire nerve, dividing it into 24 sectors of regular arrangement. With this information, the version of the program designed for specialists in Ophthalmology and Clinical Optometry separates the cup from the edge and divides it into the 6 usual angular sectors and provides much more detailed clinical information. 

Regular topographic analysis generally shows progressive cup growth, due to ischemia, especially towards the lower and upper regions. For this reason, the intermediate sectors 8 and 20 usually show a more evident reduction in hemoglobin, as the cup grows.

The analysis of sectors of the rim can be obtained in two ways: as a percentage of area of each sector in relation to the total area of the nerve (in the Excel table) or as concentrations of Hb in each sector. Glaucomatous atrophy produces a progressive reduction of the "RIM", and this can be measured by observing the percentage area of its sectors. But the hemoglobin in each sector does not have to be exactly related to the area. For example, in a well-controlled glaucomatous patient, even if there is atrophy and reduction of the RIM, it may be well perfused. Conversely, in a poorly controlled patient, perfusion is more likely to be compromised and hemoglobin is low in some sectors.

Therefore, a substantial reduction in Hb is not always found in the chart of rim and cup sectors. It should also be noted that the colors represent percentiles with respect to the normal concentration. In normal and glaucomatous cups the Hb concentration is low. Therefore, it is very frequent that the color of the cup is represented in green even if it is large and atrophic. Its size will be excessive, but its Hb concentration will probably not differ from normal.

It is most common to find defects in the lower or higher sectors of the the rim, associated with visual field defects in the opposite hemicamp, but the correspondence is usually greater with the area, while the Hb will be lower the worse controlled cases.

Progression analysis

We are currently working on the design of a progression analysis program that will analyze areas and hemoglobin concentrations separately. We will also include an index called GIP (Globin Individualized Pointer), with less diagnostic interest, but much more reproducible. This way, a better evaluation of the changes that occur in the same subject over time is achieved. This type of analysis will be incorporated in the version of the program designed for specialists in Ophthalmology and Clinical Optometry.

However, I must emphasize that the presence of Hb concentrations at normal limits is compatible with pathological GDF, since in a well-controlled patient the DL classifier can detect a cup of abnormal size and shape, vessels displaced to the nasal side or peripapillary glaucomatous atrophy, and yet no level of perfusion distinguishable from normality is observed at that time.

 

Frequency distribution in 2875 cases

To better understand what I have explained, we will represent the frequency distribution of some of these indices in 1438 normal cases and 1437 confirmed and suspected glaucomas.

As can be seen, the best ability to distinguish normals from glaucomas is obtained with the GDF.

Some areas of the nerve like 8 are affected in the glaucomas while others like 3 are not:

 

The area of the rim and the cupbetter represent the evolution phase at which the illness is …..

…. , while the Hb values of the sectors of the rim and the cup are more representative of the situation of its perfusion at a given moment.

The Deep Learning (DL) classifier between glaucoma and non-glaucoma, unlike the other indices, is not gradual, but tends to adopt extreme values (0=glaucoma and 1=normal).

One piece of information that cannot be seen in this example is the size of the disc, which is useful to know in order to interpret its influence on the relative size of the cup and rim. The estimation of this size can only be made when we have many images obtained with the same camera as well as equal amplification and resolution (width and height pixels). When uploading images of very different resolutions it is not possible to obtain the disk size and it will not be included in the results.

Prof. Manuel Gonzalez de la Rosa

Professor of Ophthalmology

Founder of Insoft SL