Using Computer Vision to Detect Surgical Instrument in Paediatric Surgery
(2026) EEML05 20261Division for Biomedical Engineering
- Abstract
- Surgical education is traditionally based on apprenticeship and supervised practice, but limited availability of experienced surgeons, restricted access to training cases, and subjective feedback create challenges for efficient surgical training. Solutions utilizing machine learning and computer vision for automating instrument tracking and automated phase recognition have been investigated in laparoscopic surgery. Limited research exists on similar solutions for open surgery and especially paediatric surgery. Our project aims to create a model capable of detecting, classifying and segmenting surgical instruments in recordings of surgeries. The dataset was created from recordings of five different surgeries performed at Skåne University... (More)
- Surgical education is traditionally based on apprenticeship and supervised practice, but limited availability of experienced surgeons, restricted access to training cases, and subjective feedback create challenges for efficient surgical training. Solutions utilizing machine learning and computer vision for automating instrument tracking and automated phase recognition have been investigated in laparoscopic surgery. Limited research exists on similar solutions for open surgery and especially paediatric surgery. Our project aims to create a model capable of detecting, classifying and segmenting surgical instruments in recordings of surgeries. The dataset was created from recordings of five different surgeries performed at Skåne University Hospitals. Thirteen classes of surgical instruments were annotated with segmentation masks. A pre-trained segmentation model was trained and evaluated using standard metrics. Most instruments were classified with high accuracy, except suture scissors, short diathermy forceps and straight clamp which could be construed to limited and inconsistent training data. The results indicate the model primarily failed by missing instruments rather than wrongly classifying them. Limitations of the study include the small dataset size, variability in recording quality, and potential bias due to all recordings originating from a limited number of surgeries and hospitals. Despite limitations, the project demonstrates that deep learning based instrument tracking in open paediatric surgery is feasible and represents a first step toward automated surgical workflow analysis. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9234087
- author
- Lundström, Svante Amadeus LU and Lindberg, Felix LU
- supervisor
-
- Maj Stenmark LU
- organization
- alternative title
- Användning av datorseende för att detektera instrument i barn- och ungdomskirurgi
- course
- EEML05 20261
- year
- 2026
- type
- M2 - Bachelor Degree
- subject
- language
- English
- id
- 9234087
- date added to LUP
- 2026-06-09 13:38:35
- date last changed
- 2026-06-09 13:38:35
@misc{9234087,
abstract = {{Surgical education is traditionally based on apprenticeship and supervised practice, but limited availability of experienced surgeons, restricted access to training cases, and subjective feedback create challenges for efficient surgical training. Solutions utilizing machine learning and computer vision for automating instrument tracking and automated phase recognition have been investigated in laparoscopic surgery. Limited research exists on similar solutions for open surgery and especially paediatric surgery. Our project aims to create a model capable of detecting, classifying and segmenting surgical instruments in recordings of surgeries. The dataset was created from recordings of five different surgeries performed at Skåne University Hospitals. Thirteen classes of surgical instruments were annotated with segmentation masks. A pre-trained segmentation model was trained and evaluated using standard metrics. Most instruments were classified with high accuracy, except suture scissors, short diathermy forceps and straight clamp which could be construed to limited and inconsistent training data. The results indicate the model primarily failed by missing instruments rather than wrongly classifying them. Limitations of the study include the small dataset size, variability in recording quality, and potential bias due to all recordings originating from a limited number of surgeries and hospitals. Despite limitations, the project demonstrates that deep learning based instrument tracking in open paediatric surgery is feasible and represents a first step toward automated surgical workflow analysis.}},
author = {{Lundström, Svante Amadeus and Lindberg, Felix}},
language = {{eng}},
note = {{Student Paper}},
title = {{Using Computer Vision to Detect Surgical Instrument in Paediatric Surgery}},
year = {{2026}},
}