Dr. Pallak Kusumgar, P16857, Dr. Rushad Shroff, Dr. Rohit Shetty
Introduction :
Keratoconus (KC) is a progressive corneal ectatic disorder characterized by stromal thinning, protrusion and the resultant irregular astigmatism and myopia.1Crosslinking of corneal collagen using a combination of riboflavin and UV-A (ultra-violet) exposure has emerged as an important management modality for progressive keratoconus2. A recent iteration of this is a nominally invasive procedure, accelerated collagen crosslinking (9 mW/cm2 and 10 minutes) which can significantly slow the progression of the disease in corneas with keratoconus3. Collagen crosslinking however, can cause an increase in corneal optical haze after the treatment and hence newer and more efficient protocols are constantly being reviewed and developed. 4 In fact, a recent report states that customized CXL is safe and leads to significant flattening of the cornea at 12 months5.
Clinicians often encounter cases where progressive nature of the disease is difficult to predict and decision regarding the management is not simple.
Artificial intelligence (AI) can play crucial role by using efficient algorithms to link topographic progression with biomechanics of soft tissue to predict the degraded zone. We present a novel AI approach to patient tomography and finite element modeling that can estimate growth of the degraded zone in progressive keratoconus and customize cross-linking based on this predicted zone.
Method
A total of 15 keratoconic eyes of 15 patients were included in this prospective study. Subjects with mild to moderate keratoconus (grades 1 and 2) upon presentation and a progressive keratometric increase by at least 1D in 6 months, intolerance to contact lens wear, thinnest pachymetry greater than 400 μm were included in this study. Patients with active allergic eye disease, active ocular inflammation, pregnancy, diabetes, glaucoma or central scarring of the cornea were excluded from the study. Detailed evaluation including refraction, anterior and posterior segment evaluation and topography was done pre-operatively for all the subjects. Topography was measured with Pentacam HR. Corneal deformation was measured with Corvis-ST. Also specular microscopy was done for all the included subjects.
A 3-D geometrical model of cornea was constructed using tomography. The model included the epithelium and stroma. To model KC, it was assumed that: (a) the disease caused local biomechanical degeneration; (b) the strength of the stroma in KC was determined by the residual collagen network that was unaffected by the disease. The model was used to compute the shape and size of the region of localized weakening in progressive keratoconus using AI and finite element method (FEM).
This three dimensional AI based model was used to design concentric treatment zones to deliver differential energy. 15 keratoconic eyes underwent topography-guided crosslinking (Avedro Inc.,USA). The topography was used to design a patient specific UV beam, which was centered and had peak intensity at a point determined by the AI model. The delivered energy varied from 15 at the center to 3 J/cm2 at the periphery of the UV beam. The maximum treated diameter was 8 mm. Patients were followed up-to 6 months post-operatively. Refraction, keratometry, lower and higher order aberrations were measured and compared pre and post surgery. Also specular microscopy was done at 6 months follow-up.
Results :
15 eyes underwent AI based customized crosslinking and were followed up for 6 months post-operatively. Refractive error significantly improved following the surgery. Mean refractive spherical equivalent reduced from -3.5± 3.1D to -2.5D± 2.6D, which was statistically significant (p=0.003). Also both spherical and cylindrical power significantly decreased from -2.1 ±2.9D to -1.7 ±2.6D (p=0.003) and -2.6 ±1.4D to -2.1 ±1.3D respectively. Root mean square of lower and higher order aberrations reduced by 42% and 44% respectively.On comparing axial and tangential maps 6 months post-operatively flattening of up-to 3.5 D was seen in the cone area with corresponding steeping of the surrounding area.
Looking at other outcomes of crosslinking, differential demarcation line with deeper line in area where more energy was delivered was seen on anterior segment OCT. Anterior stromal haze was seen more in cone area pointing towards stromal collagen compaction and remodeling.
Discussion :
Literature has postulated keratoconus to be a localized disease. Current modality of treatment, corneal cross-linking is one size fits all technique and delivers equal amount of energy all over cornea. Studies have shown customized crosslinking to be safe and effective.
In this customized method, differential energy with maximum energy being delivered to the area of maximum weakness is used. This technique uses topography to determine the zone of weakness.
In our study we propose a novel approach of artificial intelligence to predict the size and shape of weakness zone and customize our treatment on the basis of this predicted zone. This AI based customized crosslinking results in regularization of corneal surface by flattening in cone area and steeping in surrounding area.
Conclusion :
AI based customized crosslinking is an evolving novel technique in management of keratoconus. It is safe, effective and results in regularization of corneal surface with improvement in optical aberrations.
- Gordon-Shaag A, Millodot M, Shneor E, Liu Y. The genetic and environmental factors for keratoconus. BioMed research international. 2015;2015:795738.
- Arnalich-Montiel F, Alio Del Barrio JL, Alio JL. Corneal surgery in keratoconus: which type, which technique, which outcomes? Eye and vision. 2016;3:2.
- Ruberti JW, Roy AS, Roberts CJ. Corneal biomechanics and biomaterials. Annu Rev Biomed Eng. Aug 15 2011;13:269-295.
- Mazzotta C, Hafezi F, Kymionis G, et al. In Vivo Confocal Microscopy after Corneal Collagen Cross-Linking. Ocul Surf. Jun 30 2015.
- Seiler TG, Fischinger I, Koller T, Zapp D, Frueh BE, Seiler T. Customized Corneal Cross-linking: One-Year Results. American journal of ophthalmology. Jun 2016;166:14-21.
Leave a Comment