Automatic ocular version evaluation in images using random forest
Introduction
Ocular version is the name given to the joint movement of the eyes in the same direction, with both eyes open and without occlusion (Singh et al., 1973). They can be denominated supraversion, infraversion, levoversion and dextroversion when the movement is up, down, left and right, respectively.
Version examination detects and measures restrictions, paralysis and disproportionate actions of the eye muscles during the simultaneous movements of the eyes executed in the version (Wright & Spiegel, 2013). The muscles analyzed are responsible for the performance of eye movements. The examination is often used in the detection, follow-up and surgical planning of patients with strabismus.
Strabismus is a condition in which the visual axes of the eyes are not aligned (Ferri, 2014). One eye may look straight ahead, while the other eye turns inward, outward, upward, or downward. Causing the brain to receive two images at different focus, making the joining process performed by it difficult or impossible. According to Kaplan (2005) strabismus affects approximately 4% of the world’s population. Affecting mainly premature and underweight children (Rosenbaum & Santiago, 1999).
Early detection of ocular misalignment increases treatment success. Detected as a child, it allows proper visual development, avoiding problems such as low vision. An efficient treatment can also improve the visual fields and the depth perspective in adults, providing improved living conditions by opening up more significant employment and socialization (Shimauti, Pesci, Sousa, Padovani, & Schellini, 2012). A study by Hashemi et al. (2019) revealed that 1 in every 50 people had strabismus, severely affecting their quality of life.
Binocular vision disorders affect about 1–5% of the population (Carlton and Kaltenthaler, 2011, Weber et al., 2017, Hashemi et al., 2019), with comitant strabismus, the most common (about 74% of cases). Without proper treatment, patients could suffer irreversible vision loss that could be prevented (Birch, 2013), entirely preventable consequences. Thus, to decrease the effect of strabismus in adults, non-surgical solutions can be adopted, such as prism lenses and orthoptic exercises. However, in the vast majority of cases, surgery is the only solution to reduce the effect of strabismus (Ophthalmologists, 2017). In surgical treatment cases, the surgical plan is defined with the help of some exams, including the version exam.
A version test is useful for detecting “A” and “V” patterns1 associated with horizontal strabismus, mainly when it is not possible to obtain measurements and is also used for comparing the relative excursions of both eyes, for example, evaluating the patient with thyroid eye disease2 and unequal restriction of the inferior rectus muscles (Traboulsi & Utz, 2016). The version exam is used as one of the variables considered by the ophthalmic surgeon in the diagnosis and planning for strabismus surgeries. Since the version exam is, in practice, a subjective test, the creation of a method that aims to automate the examination can help in obtaining a more objective result, since there were no works in literature that propose an automatic method to quantify ocular versions.
Version test is useful for detecting “A” and “V” patterns3 associated with horizontal strabismus, mainly when it is impossible to obtain measurements. It is also used for comparing the relative excursions of both eyes, for example, evaluating the patient with thyroid eye disease4 and unequal restriction of the inferior rectus muscles (Traboulsi & Utz, 2016). The version exam is used as one of the ophthalmic surgeon’s variables in the diagnosis and planning for strabismus surgeries. Since the version exam is a subjective test in practice, creating a method that aims to automate the examination can help obtain a more objective result since no works in the literature propose quantifying the ocular versions automatically.
This paper presents a computational method to perform the automatic evaluation of ocular versions based on images, using image processing techniques, pattern recognition and computational intelligence. This method contributes the following: (1) Image acquisition protocol for automatic evaluation of the ocular version; (2) image database of version exam; (3) new method for location of the eyes in color images; (4) new sclera segmentation method; (5) new limbus segmentation method; and (6) building an automatic solution to measure the ocular version.
Regarding the following sections of this paper, Section 2 presents the related works. Section 3 describes the method steps that is used to evaluation the ocular version. In Section 4, we present and discuss the results that are achieved at each stage of the proposed methodology. Finally, Section 5 presents the conclusions, analyzing the efficiency of the techniques used and presented future works.
Section snippets
Related works
In the literature, there are works that propose the use of technology to conduct eye exams or tests, but none of them proposes using images in automatic ocular version evaluation. However, several researches have topics related to the steps of the proposed method as to how achieve a fully automated system: location of the eyes (Section 3.2.2), sclera segmentation (Section 3.2.3) and eye corner localization (Section 3.2.5).
Materials and method
This section first presents the materials used for the automatic ocular versions measurement. Then, we describe a sequence of the stages developed in order to achieve the goals of the proposed method.
Results and discussions
This section presents the results achieved in the tests performed with the proposed method for measuring eye versions using images. First, the accuracy of the method is evaluated for these steps: eye localization (Section 3.2.2), sclera segmentation (Section 3.2.3), limbus location (Section 3.2.4), eye corner location (Section 3.2.5) and automatic version measurement (Section 3.2.6). Making use of every image in the dataset for each step of the proposed method.
Finally, we performed a test of
Conclusion
This work presented an automatic method to perform version examination using image processing and pattern recognition techniques via an image of a face. The proposed method performs the version examination using eye localization, sclera segmentation, limbus localization and eye corner localization. Where locating the eyes showed solid results, achieving an accuracy of 91.49%.
The proposed method was able to segment most of the sclera in the images tested. For eye corner localization, the
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
Our research group acknowledges financial support from CNPQ (Grant No.: 423493/2016-7 and 307210/2018-9) and FAPEMA (Grant No.: UNIVERSAL-00974/18).
References (50)
- et al.
An end to end deep neural network for iris segmentation in unconstrained scenarios
Neural Networks
(2018) Amblyopia and binocular vision
Progress in Retinal and Eye Research
(2013)- et al.
Eye landmarks detection via weakly supervised learning
Pattern Recognition
(2020) - et al.
A novel iris center localization based on circle fitting using radially sampled features
- et al.
The feret database and evaluation procedure for face-recognition algorithms
Image and Vision Computing
(1998) - et al.
Efficient iris segmentation method in unconstrained environments
Pattern Recognition
(2013) - et al.
Strabismus measurements with novel video goggles
Ophthalmology
(2017) - et al.
A robust method for eye features extraction on color image
Pattern Recognition Letters
(2005) - et al.
Efficient eye corner and gaze detection for sclera recognition under relaxed imaging constraints
- et al.
A novel method for sclera recognition with images captured on-the-move and at-a-distance
Robust sclera recognition system with novel sclera segmentation and validation techniques
IEEE Transactions on Systems, Man, and Cybernetics: Systems
Computer-aided methodology for syndromic strabismus diagnosis
Journal of Digital Imaging
Segnet: A deep convolutional encoder-decoder architecture for image segmentation
IEEE Transactions on Pattern Analysis and Machine Intelligence
Amblyopia and quality of life: A systematic review
Eye
Ophthalmology review manual
Introduction to algorithms
A new efficient and adaptive sclera recognition system
Ssrbc 2016: Sclera segmentation and recognition benchmarking competition
Global and regional prevalence of strabismus: A comprehensive systematic review and meta-analysis
Strabismus
Seeing Through New Eyes: Changing the Lives of Children with Autism, Asperger Syndrome and other Developmental Disabilities Through Vision Therapy
Cited by (3)
Machine learning-based model for recognizing the failure modes of FRP-strengthened RC beams in flexure
2023, Case Studies in Construction MaterialsMultiple attentional aggregation network for handwritten Dongba character recognition
2023, Expert Systems with ApplicationsCitation Excerpt :For HDCR, the statistical features such as coarse grid (Da et al., 2011) and histogram of oriented gradient (Luo et al., 2020) are more distinguishable than certain structural features that primarily study the strokes or parts structure of Dongba characters. The most commonly used classifiers involve support vector machines (SVM) (Baldomero-Naranjo, Martinez-Merino, & Rodriguez-Chia, 2021), random forests (Pinheiro et al., 2021) and k-nearest neighbor (KNN) (Zhang, Xiao, Gao, Zhang, & Wang, 2022). Nevertheless, these algorithms are basically on the basis of a combination of independent elements and depend considerably on hyperparameters that have different optimum values in their own different datasets.
Computational method for aid in the diagnosis of sixth optic nerve palsy through digital videos
2022, Computers in Biology and MedicineCitation Excerpt :However, when much of the iris is inside the inner or outer corners, partially occluded, YOLOv3 has difficulty centering the result on the pupil. This may be because, in the training image dataset, there are not many examples of patients looking at the corners of the eye so sharply, which usually occurs in patients with strabismus, where there are version deviations and the eyes are not correctly aligned [35]. Therefore, it was decided to combine the results of eye center detection done by Isophotes and YOLOv3, forming the third tested approach for eye center detection.