Thursday, November 22, 2018

Correlation between glaucomatous visual field loss and arrangement of nerve fibers in retina and optic nerve head

Dr Syed S. Hasnain, MD
Dr Aziz Hasnain, JD

Porterville, USA




Abstract:

This presentation discusses the possible site of injury in chronic glaucoma by correlating the glaucomatous field loss with the arrangement of nerve fibers (NFs) in the retina and optic nerve head (ONH).  Glaucomatous visual field loss is a unique and diagnostic feature of glaucoma, since no other disease results in similar progressive field loss.  Glaucomatous field loss is produced in an orderly and predictable sequence which is the reason we do perimetry in glaucoma.

Glaucomatous field loss correlates fully with the arrangement of NFs while in the retina, but does not correlate at all with the arrangement of NFs after they have made their 90 degree turn into the prelaminar region.  The horizontal orientation of the NFs in the retina is changed into a vertical orientation of the NFs in the prelaminar region.  Therefore the characteristic glaucomatous field loss such as arcuate scotoma and Ronnie’s nasal step can’t be produced if the primary site of injury is in the prelaminar area or lamina cribrosa (LC) and beyond in glaucoma.

In conclusion, the nerve fibers are vertically arranged and the temporal retinal horizontal raphe is no longer present in the LC.  Furthermore, the arcuate fibers are no longer isolated and the macular fibers have become centrally placed in the LC. Therefore, the characteristic glaucomatous field loss such as arcuate scotoma and Ronnie’s nasal step can’t be produced if the LC is the site of injury in glaucoma. Although the LC is widely believed to be the primary site of injury in glaucoma, there is no correlation of glaucomatous field loss with the arrangement of NFs in the lamina cribrosa.  In view of aforementioned, the LC can’t be the primary site of injury in glaucoma.

This article suggests that the primary site of injury in glaucoma is where the retinal nerve fibers make their 90 degree turn and become prelaminar NFs.  Furthermore, the kind of injury appears to be stretching and severance of NFs at this 90 degree turn.

Arrangement of Nerve Fibers in the Retina
Before we discuss the above question, it is imperative to briefly summarize the arrangement of nerve fibers in the retina and optic nerve head (ONH) and the way visual field defects are produced in glaucoma. 

In the retina, the NFs are arranged in a characteristic way (Figures 1 & 2).  One million or so NFs are stacked in layers superficial to deep.  The NFs originating from the most peripheral retina lie deepest (closest to sclera) and exit closest to the edge of the scleral opening whereas the NFs originating closest to the optic disc lie most superficial (closest to vitreous) and exit from the most central part of the ONH.  As a result, the deepest retinal NFs make their 90 degree turn closest to the scleral edge to become prelaminar fibers, whereas the retinal NFs originating closest to the ONH lie most superficial on the anterior surface of the ONH (closest to vitreous) and make their 90 degree turn in the most central part of the ONH. 




Figure 1. Schematic Diagram: The arrangement of nerve fibers in the retina and optic disc. The most peripheral fibers (5) originate farthest from the optic disc, lie deepest (closest to sclera) and exit closest to the scleral edge. The most central fibers (1) originate closest to the disc, lie most superficial (closest to vitreous) and exit from the most central part of the disc. 


The nasal NFs enter the nasal part of the disc.  The NFs originating from the nasal macular area go directly to the temporal part of the ONH whereas the NFs originating from the temporal macular area and horizontal raphe arch above and below the macular NFs to reach the ONH.  These are the arcuate nerve fibers. (Figure 2) 



Figure 2. Schematic Diagram: The arrangement of nerve fibers in the retina. The arcuate fibers arch above and below the macular fibers to reach the poles of the optic disc.

Four parts of ONH.
1)      Retinal or superficial NFs lying horizontally
2)     Prelaminar region - after the retinal NFs have made the 90 degree turn
3)     Laminar region
4)     Retrolaminar region

The 360 degrees of retinal NFs converge on the ONH as a superficial surface layer and are arranged on the anterior surface of the ONH in the same way as in the retina.  After making the 90 degree turn, the retinal NFs become prelaminar NFs.

In the prelaminar region, the loose NFs begin to get arranged in bundles. The macular NFs start shifting to occupy the central position, thus the arcuate fibers lose their arcuate pattern.  The arcuate NFs get mixed with the rest of the temporal fibers and become indistinguishable.  Moreover, the NFs in the prelaminar region become vertically oriented and thus the temporal retinal horizontal raphe disappears.  In the lamina cribrosa, the bundles of NFs get fastened in its pores and the arrangement of NFs becomes drastically different in the LC compared to while the NFs were in the retina.  In the laminar region, the arcuate fibers are no longer distinguishable and the horizontal raphe also disappears.   

Glaucomatous Visual Fields
Glaucomatous field defects are produced in an orderly and predictable sequence which is the basis for perimetry in glaucoma.  In glaucoma, the peripheral NFs are destroyed first resulting in peripheral field constriction.  But the peripheral field loss has poor diagnostic value since many other diseases such as cataract can also result in peripheral field loss.  In the early stages of glaucoma along with peripheral field constriction, the isolated scotomas begin to appear in both superior and inferior paracentral areas (10 to 20 degrees) which are very diagnostic of glaucoma.  As glaucoma progresses, the isolated scotomas become more frequent and ultimately coalesce to form sharply-defined double arcuate field defects.

The superior arcuate scotoma usually appears first.  Both superior and inferior arcuate scotomas start from the blind spot and end sharply at the horizontal nasal hemifield giving rise to Ronnie’s nasal step.  Meanwhile the peripheral field loss is also progressing towards the center and ultimately joins the double arcuate scotoma.  When this occurs, only about 10 degrees of central vision remains which ultimately is also lost and a subject becomes 100% blind.

1.  Can the glaucomatous field defects be produced if the lamina cribrosa is the primary site of injury in glaucoma?

Unlikely.  Although the lamina cribrosa is widely believed to be the primary site of injury in glaucoma, the glaucomatous field defects contradict LC as the site of injury.  It is theorized that elevated IOP causes posterior bowing of the LC (cupping) resulting in distortion and misalignment of its pores, thereby impeding the axoplasmic transport leading to death of RGCs.  But this theory can’t be valid in context of the orderly loss of NFs in glaucoma.  The following are arguments against LC being the site of injury:

1)  It is inconceivable that the multilayered rigid connective tissue LC, densely packed with NFs is so fragile that it will start bowing posteriorly with an elevation of only 5-10 mmHg of IOP, yet it wouldn’t bow in cases of acute glaucoma in which the IOP becomes extremely elevated. There is no acute cupping occurring in acute glaucoma.

2)  There is no histology supporting the posterior bowing of LC.  Instead, we have confirmed evidence of posterior migration of the LC from the early stages of glaucoma.  It is inconceivable that a loosened and detached LC while migrating posteriorly could also become posteriorly bowed, especially with elevation of only 5-10 mmHg of IOP.

3) If the LC was indeed bowing posteriorly, then its central part should be affected first resulting in the loss of central vision, initially.  Thus, there should be doughnut-shaped field defects.  But in actuality, peripheral vision is destroyed first and central vision remains until the end-stage in glaucoma. 

4) Most importantly, the glaucomatous field defects closely follow the arrangement of NFs while in the retina.  But after making the 90 degree turn, the NFs become vertically oriented in the prelaminar region and in the lamina cribrosa.  In the LC, neither the arcuate fibers are isolated nor is the horizontal raphe present as they were in the retina.

In view of the re-orientation of NFs in the lamina cribrosa, the glaucomatous field loss such as the sharply-defined arcuate field defects and Ronnie’s nasal step can’t be produced if LC is the primary site of injury in glaucoma.  In view of the aforementioned, the LC can’t be the site of injury in glaucoma.

2. Can the glaucomatous field defects be produced if retina was the primary site of injury? 

Unlikely.  Although glaucomatous field defects conform to the arrangement of nerve fibers in the retina, it is inconceivable that elevated IOP would destroy the retinal NFs in an orderly fashion from peripheral to central and not destroy the entire retinal NFs at once.  In glaucoma, the NFs are being destroyed in an orderly sequence spreading over several years staring with the most peripheral NFs and ending with the central fibers.  Therefore, the retina itself can’t be the primary site of injury in glaucoma.

We face a dilemma.  We have ruled out that the nerve fibers while in the retina or even as a superficial layer of the ONH can’t be the site of injury, although the glaucomatous field loss fully correlates with the arrangement of retinal NFs.  Furthermore, we have also ruled out that the prelaminar and laminar region also can’t be the site of injury in glaucoma because of the lack of correlation of glaucomatous field loss with the arrangement of NFs in these regions.  Then, why does the characteristic glaucomatous field loss correlate with the arrangement of nerve fibers in the retina?  

3. Can the scleral edge be the primary site of injury in glaucoma?

We are only left with the area where the retinal NFs cross the edge of the scleral opening and make a 90 degrees turn into prelaminar region.  After the retinal NFs cross the scleral edge they are supported underneath by the lamina cribrosa. 
The LC is firmly kept in place in the scleral opening by the border tissue of Elschnig (BT) and also by retinal NFs as roots anchor a tree.  In normal circumstances, the LC provides a good underneath support to the NFs during lifetime. However, if the BT degenerates and the LC starts sinking, then serious consequences can occur to the NFs after they cross the scleral edge.

Glaucoma:  A Two-Stage Disease.
It is hypothesized that glaucoma is a two-stage disease. The first being a biological stage, followed by a second, mechanical stage.

The First (Biological) Stage
In the biological stage, there is degeneration of the border tissue of Elschnig due to chronic ischemia resulting from the chronic compression of BT circulation by elevated IOP.  Ciliary perfusion pressure supplying the BT and IOP are opposing forces.  Normally, the ciliary circulatory pressure supplying the BT should be higher than IOP for its good perfusion and healthy maintenance.  However, if this healthy relationship is reversed either due to a rise in IOP (ocular problem) or if perfusion pressure of the BT becomes lower than the IOP due to conditions such as chronic hypotension (systemic problem), then even the normal range IOP will take the upper hand and act as an elevated IOP.  This would lead to chronic ischemia and degeneration of the BT, thus resulting in normal-tension glaucoma in the systemically compromised subjects (Figure 3).



Figure 3. Graphic diagram: The interaction between ciliary pressure and IOP. Normally, the ciliary pressure supplying the border tissue should be higher than IOP for its good perfusion and healthy maintenance as in column (A). In column (B) the IOP is increased to 30 mmHg due to an ocular problem whereas the ciliary pressure is still the same at 25, resulting in high-tension glaucoma. In column (C) the ciliary pressure is decreased to15 due to problems such as systemic hypotension so even normal IOP at 20 will act as an elevated IOP in this scenario so resulting in normal-tension glaucoma.

Due to degeneration of BT, the lamina cribrosa becomes loose and starts sinking in the scleral canal.  This is the beginning of the second or mechanical stage.  This may also be called perimetric glaucoma.

The Second (Mechanical) Stage:
Due to sinking of the LC, the most peripheral and deepest retinal NFs (being closest to the scleral edge) will be stretched and broken against the scleral edge first (Figure 4, NF 5).


Figure 4. Schematic diagram: Due to sinking of the LC the most peripheral and deepest prelaminar nerve fibers (5) are stretched and severed against the scleral edge first. The next-in-line fiber (4) will move towards the scleral edge and also gets severed. This process will continue in an orderly sequence until the most central fiber (1) is severed.

The next in line fiber (NF 4) will move towards the edge to occupy the space vacated by the preceding severed fiber and will also get stretched and broken.  This process will continue in an orderly sequence until the most central fiber has moved towards the scleral edge and gets severed.

The actual site of injury appears to be when the retinal NFs make the 90 degree turn into the prelaminar region.  The sinking of the LC will result in loss of underlying support and the loss of continuity of nerve fibers.  This scenario would result in stretching and severance of the NFs. The severance of NFs would result in further sinking of LC as NFs are anchoring the LC as roots anchor a tree.  This process will become self-propagated until all the NFs are severed and gone. This is probably the reason glaucoma can’t be halted despite maximal lowering of IOP.  Glaucoma may not be an optic neuropathy but an optic disc axotomy.

Conclusion
Glaucomatous field defects such as the arcuate scotoma and Ronnie’s nasal step correlate fully with the arrangement of nerve fibers while in the retina or before the nerve fibers make the 90 degree turn into the prelaminar region. Therefore the site of injury to the NFs has to be either at the 90 degree turn or before but not after the NFs have made the 90 degree turn into the prelaminar region.

In the prelaminar region, the NFs become vertically oriented and the macular NFs start shifting to the central part.  The arcuate NFs get mingled with the rest of the temporal fibers and lose their arcuate pattern.  Furthermore, the loose NFs begin to form bundles in the prelaminar region which become fastened in the pores of the LC so it would be unlikely any injury could produce minute isolated paracentral scotomas which ultimately coalesce to form the arcuate scotoma. 

In the laminar region the NFs become vertically oriented, the arcuate NFs are no longer isolated and the horizontal raphe also disappears.  In other words, the arrangement of NFs in the LC is drastically different than the arrangement of NFs in the retina.  Glaucomatous field loss such as the arcuate scotoma and Ronnie’s nasal step can’t be produced if the primary site of injury is in the lamina cribrosa. If the LC was the site of injury then we should have seen doughnut-shaped field defects due to loss of central vision initially.

We have ruled out not only the prelaminar and laminar region but also the retina as the primary site of injury in glaucoma.  Thus, we are left with the area where the NFs cross the scleral edge on to the LC.  This is a weak and vulnerable area where the NFs cross the scleral edge and enter into the prelaminar area.  In normal circumstances, the LC plate provides a good underneath support to the NFs. However, due to sinking of the LC the NFs lose their firm underneath support which is important for their 90 degree angulation and continuity.  As a result of sinking of the LC, the NFs are stretched and severed against the scleral edge.

Analogy: a road is made of NFs which converge on the manhole cover in the middle of a road.  If the manhole cover begins to sink due to deterioration of its border area, then the road NFs due to loss of their continuity, will be stretched and broken at the edge.  The similar phenomenon appears to be occurring in the glaucomatous disc. The sinking of the LC and the severance of NFs will become self-propagated and would continue until all the NFs are severed in an orderly sequence from peripheral to central in glaucoma.

In summary, the site of injury to the NFs in glaucoma appears to be where the NFs are making a 90 degree turn.  This will still produce glaucomatous field defects correlating with the arrangement of NFs while they are in the retina.  Once the NFs have made the 90 degree turn into prelaminar region, their arrangement changes drastically therefore any injury to NFs in the prelaminar region and lamina cribrosa  cannot produce the classical glaucomatous field defects. In view of the aforementioned, the lamina cribrosa can’t be the site of injury in glaucoma. Furthermore, due to severance of NFs, glaucoma may not be an optic neuropathy but an optic axotomy.



REFERENCES

1. Hasnain SS. Scleral edge, not optic disc or retina is the primary site of injury in chronic glaucoma. Medical Hypothesis 2006; 67(6) ;1320-1325
2. Hasnain SS. Scleral edge, not optic disc or retina is the primary site of injury in chronic glaucoma. Medical Hypothesis 2006; 67(6) ;1320-1325
3. Hasnain SS. Optic Disc may be Sinking in Chronic glaucoma. Ophthalmology Update. Pakistan Oct-Dec. 2010; 8 (4); 22-28.
4. Hasnain SS. Pathogenesis of Arcuate Field Defects in Glaucoma. Highlights of Ophthalmology, Panama 201240(6)
5. Yang H. et al. Posterior (outward) migration of the lamina cribrosa and early cupping in monkey experimental glaucoma. Invest Ophthalmol Vis Sci 2011;52:7109-21 11.
6. Yang H. Optic Nerve Head(ONH) Lamina Cribrosa Insertion Migration and Pialization in Early Non-Human Primate Experimental Glaucoma. Poster Presentation ARVO meeting May 03, 2010.

Sunday, October 21, 2018

AN INTRODUCTION TO GLAUCOMA

EDITORIAL IN THE OPEN OPHTHALMOLOGY JOURNAL






Monday, October 1, 2018

TOOLS IN ARTIFICIAL INTELLIGENCE



Computer systems are far from perfect. Computer glitches, loss of data, insufficient speed, hacking and other shortcomings continue to afflict these systems. Therefore, Artificial Intelligence (AI) dependent tools are being developed to overcome these problems. Some of the AI tools are as follows: 
  1. Search and optimization
  2. Logic
  3. Probabalistic methods for uncertain reasoning
  4. Classifiers and statistical learning methods
  5. Artificial Neural Networks (ANN)

 (1) Search & optimization: One way to tackle problems affecting computerized systems and AI is by performing an intelligent search of possible solutions. A “search algorithm” is any algorithm which solves the problem involving searches. In other words, it is able to retrieve information stored within some data structure or calculated in the search space of the problem domain. Examples include search engines such as Google and Yahoo, which are searchable data banks. In computer lingo, “Search Space or State Space” is the number of places to search or the space containing all feasible solutions. Some examples of search algorithms include: linked list, array data structure and search tree. Search algorithms are used for a number of AI tasks including “path-finding”. Pathfinding or pathing is the plotting, by a computer application, of the shortest route between two points.

When the space search grows to astronomical proportions due to vast amounts of data, the search becomes too slow or never completes (information explosion). Therefore, “heuristics” (or rules of thumb) are used to prioritize choices in favor of those that are more likely to reach a goal and to do so in a shorter number of steps. Heuristics can also entirely eliminate some choices that are unlikely to lead to a goal (“pruning the search tree”). Heuristics reduce the sample size and supply a program with the “best guess” for the path which leads to the solution.

For example, if the search is done for the word “fundus”, the search engine may turn up 82,50,000 results, including those for fundus of the eye, fundus of uterus, fundus of gall bladder and so on. A search engine may also advise to refine the search. This will avoid the search engine from slowing down or its inability to trace the required item. Similarly, when we use a GPS application to reach a destination, the program will find the best route, eliminating unnecessary turns and traffic jams from the search.


In “optimization”, the search starts with some form of guess and the guess is refined incrementally until no more refinements can be made. Some forms of optimization include: hill-climbing, simulated annealing, beam search and random optimization.


Swarm intelligence (SI) is also used for optimization. It is the collective behavior of decentralized, self-organized systems, whether natural or artificial. SI systems consist typically of a population of simple agents or boids interacting locally with one another and with their environment. The agents in a SI system follow very simple rules. However, when these agents interact with each other, it leads to the emergence of “intelligent” global (or collective) behavior, unknown to each individual agent. Examples of SI include: ant-colonies, bird-flocking, animal-herding, bacterial-growth, fish-schooling and microbial-intelligence.


 (2) Logic= Logic is used for knowledge representation and problem solving. Different forms of logic are utilized in AI research. 

Propositional logic involves truth functions such as “or” and “not”. First order logic adds quantifiers and predicates, and can express facts about objects, their properties and their relations with each other.   

Another concept is fuzzy sets (aka uncertain sets) which are somewhat like sets whose elements have degrees of membership (i.e. relationship with each other). Fuzzy set theory assigns a “degree of truth” (between 0 and 1) to vague statements that are too linguistically imprecise to be completely true or false (Fuzzy= vague, indistinct). Fuzzy logic is successfully used in control systems to allow experts to contribute vague rules that can be numerically refined within the system. However, Fuzzy logic doesn’t scale well in knowledge bases; many AI researchers question the validity of chaining fuzzy-logic inferences. Qualitative symbolic logic is brittle and scores poorly in the presence of noise or other uncertainty. It is also difficult for logical systems to function in the presence of contradictory rules.

(3) Probabilistic methods for uncertain reasoning= A number of problems in AI (involving reasons, planning, learning, perception and robotics) require the agent to operate with incomplete or uncertain information. Bayesian networks are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference logarithm), learning (using the expectation-maximization algorithm), planning (using decision networks) and perception (using dynamic Bayesian networks).

A Bayesian network, Bayes network, belief network, Bayes(ian) model or probabilistic directed acyclic graphical model is a probabilistic graphical model (a type of statistical model) that represents a set of variables and their conditional dependencies via a directed acyclic graph (DAG). Bayesian networks are probabilistic because they are built from probability distributions and also use the laws of probability for prediction and anomaly detection, for reasoning and diagnostics, decision making under uncertainty and time series prediction.

Bayesian networks can be used to build models from data and/or expert opinion.

A Bayesian network is a graph made up of Nodes and directed Links between them. This network, consisting of nodes and links, is regarded as a “structural specification”. Each Node represents a variable, such as height, age or gender. Links are added between nodes to indicate that one node directly influences the other.

(4) Classifiers and statistical learning methods= A classifier is an algorithm that maps the input data to a specific category. Classifiers are a function that uses pattern matching to determine a closest match. They can be tuned according to “examples”. These examples are known as observations or patterns. For example, when you receive multiple emails, some of them go to the junk folder as they are classified as useless depending on your preference for opening emails. A classifier can be trained using statistical and machine learning approaches. The most widely used machine learning algorithm is the “Decision Tree”. Other Classifiers are: neural network, k-nearest neighbor algorithm, kernel methods such as support vector machine (SVM), Gaussian mixture model and naive Bayesian Classifier. Model based Classifiers perform well if the assumed model is an extremely good fit for the actual data. If no matching model is available, and if accuracy is the sole concern, discriminative classifiers (especially SVM) tend to be more accurate.

Decision Tree

(5) Artificial Neural Networks (ANN) = ANN is defined as: “a computing system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs”. Neural networks or neural nets were inspired by the architecture of neurons in the human brain. ANNs are composed of multiple nodes, which imitate biological neurons of the human brain. The neurons are connected by links and interact with each other. ANN has made possible computers "to think and understand the world in the way humans do, while retaining the innate advantages they hold over us, such as, speed, accuracy and lack of bias". Data fed to an ANN is able to make statements, decisions or predictions with a degree of certainty.
A simple “neuron” N accepts inputs from multiple other neurons, each of which, when activated (or “fired”), cast a weighted “vote” for and against whether neuron N should itself activate. (In other words, whether the information should be further propagated). Learning requires an algorithm to adjust these weights based on the training data (By training the network, it can be decided whether the information should be further propagated). Each link is associated with the “weight”. The net forms “concepts” that are distributed among a subnetwork of shared neurons that tend to fire together.
There are 2 types of ANN:
(a) Feedforward: Here the information flow is unidirectional. A unit sends information to another unit from which it does not receive any information. There are no feedback loops. They are used in pattern generation/recognition/classification and have fixed inputs and outputs.
(b) Feedback ANN: Here feedback loops are allowed. They are used in content addressable memories.
If the network generates a “good or desired” output, there is no need to adjust the weights. However, if the network generates a “poor or undesired” output or an error, then the system alters the weights in order to improve subsequent results.

Feedforward ANN
 
Feedback ANN

Machine Learning in ANNs:
ANNs are capable of learning and they need to be trained. There are several learning strategies=
A. Supervised learning: Here labeled data is used to train algorithms. Algorithms are trained using marked data, where the input and output are known. The data is input in the algorithm, which is known as “Features”. The output is then matched between actual productions with expected correct outputs to find errors. The model can then be modified accordingly.
B. Unsupervised learning: It is required when there is no example data set with known answers. For example: searching for a hidden pattern. Unlabeled data is used to train the algorithm. The purpose is to explore the data and find some structure within it.
C. Reinforcement learning: This strategy is built on observation. The ANN makes a decision by observing its environment. If the observation is negative, the network adjusts its weights to be able to make a different required decision the next time.
Convolutional Neural Network (CNN or ConvNet)= It is a class of deep, feed-forward, ANN. It is usually applied in the field of visual imagery. They are also known as Shift Invariant or Space Invariant Artificial Neural Networks (SIANN), based on their shared-weights architecture and translation variance characteristics.
CNNs were inspired by biological processes in that the connectivity pattern between neurons resembles the organization of the animal visual cortex. CNNs use a variation of multilayer perceptrons, designed in such a way that they require minimal preprocessing. Unlike ANN, which receives simple information (vectors), the input in CNN is a multi-channeled image.

A CNN consists of an input and output layer, as well as multiple hidden layers. The hidden layers of CNN typically consist of convolutional layers, pooling layers, fully connected layers and normalization layers.
Convolutional layers apply a convolutional operation to the input, passing the result to the next layer. The convolution mimics the response of an individual neuron to visual stimuli.

Architecture of CNN
The picture above shows a CNN coming up with the best response in identifying the image, that is, dog.

Deep Learning: Also known as Deep Structured Learning or Hierarchical Learning. It is a part of AI concerned with mimicking the learning approach used by humans to gain certain types of knowledge.

Traditional Machine-Learning algorithms are linear while Deep-Learning algorithms are stacked in hierarchy of increasing complexity and abstraction. Information passes through multiple hidden layers , thus the process is called "deep". In deep learning the program builds the "feature set" by itself without supervision. Unsupervised learning as seen in deep learning is faster and more accurate.