Could AI-Based Real-Time Surgical Guidance Reduce Complications During Cataract Surgery?


A recent study suggests that a conceptual deep neural network-based surgical guidance platform is accurate and has potential as a valuable tool in phacoemulsification procedures.

Despite advances in surgical technique and instrumentation, surgeons performing phacoemulsification procedures must be concerned with variables that can adversely affect the safety of the procedure and patients’ visual outcomes. A study published in JAMA Ophthalmology investigated the potential of an artificial intelligence (AI)-based surgical feedback platform to assist surgeons and improve procedures.

Phacoemulsification is a cataract procedure in which a surgeon emulsifies the inner lens of the eye, suctions the lens out of the eye, and then replaces it with a balanced saline solution to preserve the anterior chamber of the eye. Variables such as rapid changes in intracameral flow and associated fluid turbulence, instrument placement, and visualization of the intraocular tissue can affect the performance and results of the procedure.

Previously, computer vision algorithms and deep neural networks (DNNs), a type of AI, were used for postoperative analysis. They have been shown to be able to segment the pupil and surgical instruments, analyze features to determine a trainee’s performance, and identify surgical stages in pre-recorded videos. Various convolutional neural networks (CNNs) were up to 95% accurate, according to the study authors.

For this study, researchers developed a surgical guidance platform using a region-based CNN (R-CNN)—a subtype of DNN—and evaluated the platform’s ability to track the pupil, identify surgical stages, and activate certain computer vision tools to potentially assist surgeons with real-time audiovisual feedback during phacoemulsification surgery.

Ten surgical procedures performed by treating and prospective physicians at the University of Illinois Hospital and Health Sciences Center were recorded frame-by-frame using a stereoscopic surgical microscope, and 6 were randomly selected to use the DNN in the following feedback and visualization tools schools:

  • Capsulorhexis guidance for improved symmetry and intended size of the rhexis
  • Feedback on eye decentering, erratic tool movement and turbulent flow conditions
  • Improved visualization of anatomical structures through contrast compensation, including visualization of the rhexis, residual lens fragments, and cortical fibers

The main outcomes were the area under the receiver operator characteristic curve (AUROC) and the area under the precision recall curve (AUPR) for surgical phase classification and the Dice score for pupillary boundary detection.

The DNN was trained with 600 frames and evaluated with 23,640 frames in a heterogeneous group of phacoemulsification cases. It was compared to 10,100 frames from the publicly available Cataract-101 dataset to assess the generalizability of the platform.

The DNN achieved AUROC values ​​of 0.996 for capsulorhexis, 0.972 for phacoemulsification, 0.997 for cortex removal and 0.880 for idle phase detection, researchers found. When applied to the external data set, the average drop in performance was 6.8%. The final algorithm achieved a Dice score of 90.23% for student segmentation in the local dataset and 85.4% in the external dataset. The final platform had a processing rate of and an average processing speed of 97 (standard deviation 34) frames per second.

Eleven cataract surgeons evaluated the surgical guidance platform post hoc, of which 8 (72%) responded that they would use the tool mainly or very likely for complex procedures. Five (45%) of respondents indicated that they would find it useful for non-complex procedures. All participants agreed that the platform could be useful for real-time surgical guidance during phacoemulsification procedures, and 10 out of 11 participants considered the pupil tracking and phase classification tools to be mostly or extremely accurate.

The results were also obtained with a relatively small number of frames to train the DNN.

“This proof-of-concept investigation suggests that a pipeline from a surgical microscope could be integrated with neural networks and computer vision tools to provide real-time surgical guidance,” the authors concluded, noting that the Feasibility of implementing such a system requires more research to determine.


Nespolo RG, Yi D, Cole E, Valikodath N, Luciano C, and Leiderman YI. Evaluation of artificial intelligence-based intraoperative guidance tools for phacoemulsification cataract surgery. JAMA Ophthalmol. Published online January 13, 2022. doi:10.1001/jamaophthalmol.2021.5742


Comments are closed.