Page 51 - Read Online

Page 455 Goyal et al. Art Int Surg. 2025;5:450-64 https://dx.doi.org/10.20517/ais.2024.108

Table 1. Baseline characteristics of included studies in the study
Video Inguinal hernia Number
Authors Year of Country collection procedure Laparoscopic/robotic of Single Unilateral/bilateral
surgeons centric/multicentric hernia
time
publication
period (TAPP/TEP/RIHR) involved
Ortenzi et 2023 USA October TEP Laparoscopic 85 Multicentric Yes (270 unilateral;
al. [10] 25, 2019, 349 bilateral cases)
and
December
13, 2022
Takeuchi 2023 France July 2019 TAPP Laparoscopic 11 Single Unilateral
[20]
et al. and April
2021
Choksi et 2023 USA - RIHR Robotic 8 Single Unilateral
[21]
al.
Takeuchi 2022 France July 2019- TAPP Laparoscopic 11 Single Yes (86 unilateral; 33
[8]
et al. February bilateral)
2021
Zang et 2023 USA Spring RALIHR (robotic Robotic 8 Single Unilateral
al. [9] 2022 assisted)
Zygomalas 2024 Greece - TAPP Laparoscopic 5 Single Unilateral
et al. [7]
TAPP: Transabdominal preperitoneal; TEP: totally extraperitoneal; RIHR: robotic inguinal hernia repair.

Table 2. Pre-processing and number of videos analyzed in included studies
For Internal
Authors Pre-processing of videos Total Test
training validation
[10]
Ortenzi et al. Surgical Intelligence platform used for automatic video capture, de-identification, 619 371 93 155
and cloud upload
[20]
Takeuchi et al. Uniform video sampling to prevent selection bias; ~18.12 ± 11.59 images extracted 160 130 - 30
per video. 2,356 training images; 300 test images representing the CVMPO
visual inspection phase
Choksi et al. [21] Videos down-sampled to 1 frame per second (FPS) to reduce memory usage for 211 106 30 60
phase segmentation
Takeuchi et al. [8] Videos annotated and organized using the Indexity Annotation Platform at 119 89 - 30
IRCAD France and IHU Strasbourg
[9]
Zang et al. Frame rate reduced from 30 FPS to 1 FPS using the FFmpeg library to minimize 209 188 - 21
computational load
[7] * *
Zygomalas et al. 25 TAPP videos processed to extract 1,095 images, annotated using Bounding 25 800 - 295
Boxes (11 object classes)
*
Zygomalas et al. randomly selected images from 25 study videos for training and testing the AI model. AI: Artificial intelligence.

[Table 2]. These pre-processing steps ensured consistent, high-quality data input for subsequent annotation
and model development.

Annotation software
The term annotation refers to the process of manually marking the start and end points of each procedural
step. The included studies utilized various annotation methods and software to process the videos prior to
testing and internal validation. Manual annotation was employed in multiple studies, including those
[21]
[10]
[9]
conducted by Ortenzi et al. (2023) , Choksi et al. (2023) , and Zang et al. (2023) . Computer Vision
Annotation Tool (CVAT) was specifically utilized in the study by Takeuchi et al. (2023)[20], while Indexity
[8]
Annotation Platform was adopted in a separate study by Takeuchi et al. (2022) . In contrast,
[7]
SuperAnnotate v1.1.0 was used in the study by Zygomalas et al. (2024) . These annotation tools facilitated
detailed video labeling, which is critical for creating training, validation, and testing datasets essential for