August 2019
Volume 60, Issue 11
Open Access
ARVO Imaging in the Eye Conference Abstract  |   August 2019
Predicting the likelihood of future keratoplasty from imaging corneal parameters using manifold learning
Author Affiliations & Notes
  • Siamak Yousefi
    Ophthalmology, University of Tennessee Health Science Center, Memphis, Tennessee, United States
  • hidenori takahashi
    Ophthalmology, Jichi Medical University, Japan
  • Takahiko Hayashi
    Ophthalmology, Yokohama Minami Kyosai Hospital, Japan
  • Hironobu Tampo
    Ophthalmology, Jichi Medical University, Japan
  • Satoru Inoda
    Ophthalmology, Jichi Medical University, Japan
  • Yusuke Arai
    Ophthalmology, Jichi Medical University, Japan
  • Penny Asbell
    Ophthalmology, University of Tennessee Health Science Center, Memphis, Tennessee, United States
  • Footnotes
    Commercial Relationships   Siamak Yousefi, None; hidenori takahashi, None; Takahiko Hayashi, None; Hironobu Tampo, None; Satoru Inoda, None; Yusuke Arai, None; Penny Asbell, None
  • Footnotes
    Support  Partly through unrestricted grant from Research to Prevent Blindness (New York, NY)
Investigative Ophthalmology & Visual Science August 2019, Vol.60, PB0141. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Siamak Yousefi, hidenori takahashi, Takahiko Hayashi, Hironobu Tampo, Satoru Inoda, Yusuke Arai, Penny Asbell; Predicting the likelihood of future keratoplasty from imaging corneal parameters using manifold learning. Invest. Ophthalmol. Vis. Sci. 2019;60(11):PB0141.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : To determine if machine learning can screen patients for ectatic disease and to determine if corneal shape, thickness, and elevation parameters can be used to identify patients who may be at higher risk for penetrating keratoplasty (PKP), lamellar keratoplasty (LKP), descemet’s stripping automated endothelial keratoplasty (DSAEK) or descemet’s membrane endothelial keratoplasty (DMEK) intervention

Methods : We selected 3,318 corneal optical coherence tomography (OCT) images from the baseline visit of 12,242 eyes (Casia instrument, Tomey, Japan). We applied principal component analysis on 424 corneal shape, thickness, and elevation parameters followed by t-distributed stochastic neighbor embedding (tSNE) manifold learning to reduce the number of dimensions. We employed density-based unsupervised clustering. Our post hoc analysis revealed that clusters were attributed to different keratoconus stages. Assessment of 333 eyes with post-operative keratoplasty showed that the method can also predict the risk for future keratoplasty.

Results : The mean age of participants was 69.7 (SD=16.1) and 59% were female. Manifold learning of 18 principal components followed by unsupervised clustering identified five non-overlapping clusters (Fig. 1). Clusters 1 and 2 composed of mainly normal eyes and eyes with small ectasia (assessed by ectasia screening index; ESI of Casia, Fig. 2). Cluster 3 corresponded to early stage anterior and moderate stage posterior ectasia. Cluster 4 attributed to early stage posterior but moderate stage anterior ectasia. Cluster 5 was corresponded to advanced stage anterior and posterior ectasia. The specificity was 96.5% and the sensitivity was 90.4%. The normalized likelihood of future surgery for eyes mapped onto clusters 1 to 5 were 3.1%, 2.9%, 45.3%, 29.5%, and 19.2%, respectively.

Conclusions : Manifold learning can identify the ectasia status from corneal parameters and assist the refractive surgeon in identifying those patients who may be at higher risk for future PKP and LKP. Moreover, the method can identify patients who may be at higher risk for future DSAEK and DMEK interventions.

This abstract was presented at the 2019 ARVO Imaging in the Eye Conference, held in Vancouver, Canada, April 26-27, 2019.

 

Figure 1. Left: Corneal data on tSNE map, middle: clusters, and right: predicting the likelihood of future keratoplasty (eyes with post-operative keratoplasty marked with asterisk).

Figure 1. Left: Corneal data on tSNE map, middle: clusters, and right: predicting the likelihood of future keratoplasty (eyes with post-operative keratoplasty marked with asterisk).

 

Figure 2. Left: Total ESI index, Middle: Anterior ESI index, Right: posterior ESI index of eyes at each cluster.

Figure 2. Left: Total ESI index, Middle: Anterior ESI index, Right: posterior ESI index of eyes at each cluster.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×