SICS-105: Phase recognition in Manual Small-Incision Cataract Surgery

This prospective cross-sectional study introduces the first Manual Small-Incision Cataract Surgery (SICS) video dataset, which is prevalent but understudied in low- and middle-income countries (LMICs), evaluates effectiveness of phase recognition through deep learning (DL) using the MS-TCN++ architecture and compares its results with the well-studied phacoemulsification procedure using the Cataract-101 public dataset. Our novel SICS-105 dataset involved 105 patients recruited at Sankara Eye Hospital in India. Performance is evaluated with frame-wise accuracy, edit distance, F1-score, Precision-Recall AUC, sensitivity, and specificity. The MS-TCN++ architecture performs better on the Cataract-101 dataset across reported metrics, with an accuracy of 89.97% [CI 86.69-93.46%] compared to 84.94% [CI 79.45-92.03%] on the SICS-105 dataset (ROC AUC 99.10% [98.34-99.51%] vs. 98.22% [97.17-99.39%]). Reducing the 20 phases to 13 phase in SICS improved performance without completely bridging the gap. The per-video accuracy distribution and confidence-intervals overlap between the samples. PR-AUC curves for each phase in the SICS dataset range from 46.20 to 94.18%. In conclusion, we replicated phase recognition results from phacoemulsification in a new open-access SICS dataset using DL with slightly lower prediction performance. This research marks a crucial step towards improving postoperative analysis and training for SICS.

SICS-105 is a subset of our more recent SICS-155 dataset: https://doi.org/10.5281/zenodo.15044589