Rectification of Cornea Induced Distortions in Microscopic Images for Assisted Ophthalmic Surgery

Area of Science:

• Ophthalmic surgery
• Medical imaging
• Computer vision

Background:

• Ophthalmic surgery requires accurate geometric information from intra-operative sensor data.
• High refractive power of the eye causes image distortions, challenging geometric accuracy.
• Cornea-induced distortions (CIDs) in anterior eye images are a key challenge.

Purpose of the Study:

• To develop a method for correcting cornea-induced distortions (CIDs) in surgical microscope images.
• To improve geometric accuracy for computer-assisted ophthalmic surgery systems.
• To introduce a synthetic dataset for training and evaluating CID correction models.

Main Methods:

• A convolutional neural network (CNN) with stereo fusion layers was used to predict distortion distribution maps (DDMs).
• A synthetic dataset, CIDCAT, was generated using a digital eye model for supervised learning.
• Domain regularization via semantic segmentation was employed to bridge the gap between synthetic and real surgical images.

Main Results:

• The rectification model reduced cornea-induced pupil radius error from 8.56% to 0.72%.
• Structural similarity in images improved by over 9% for synthetic data.
• The domain regularization technique facilitated successful application to real surgical images.

Conclusions:

• The CIDCAT dataset and rectification model enable CID investigation and correction.
• The proposed model effectively minimizes CIDs while maintaining image integrity.
• This work advances computer-assisted and robotic ophthalmic surgery by enabling distortion-free 3D eye understanding.