Sciuto, Antonio (2023) Machine Learning for Health. A Case Study in Colorectal Surgery. [Tesi di dottorato]

Anteprima

Testo Sciuto_Antonio_36.pdf Download (13MB) | Anteprima

Abstract

Colorectal diseases including cancer and several benign conditions lead to a significant number of surgical procedures worldwide. Today, these surgeries are commonly performed using a minimally invasive approach, such as laparoscopy or robot-assisted surgery. Reduced blood flow to the remaining healthy bowel is associated with significant postoperative complications, which negatively affect patient outcomes and increase healthcare costs. In recent years, near-infrared imaging using indocyanine green fluorescence has become a promising tool for intraoperative assessment of bowel perfusion. Temporal trend of fluorescence rather than its intensity may better predict the onset of complications. However, one major limitation of this technique is that fluorescence evaluation relies on the surgeon’s subjective interpretation. Intraoperative decisions based on indocyanine green fluorescence angiography are influenced by the surgeon’s experience and exhibit a considerable interindividual variation. Artificial intelligence is increasingly being explored as a powerful instrument in the surgical field to improve precision and ultimately enhance operative performance and patient outcomes. Due to its subjective nature, fluorescence assessment of bowel perfusion may benefit from the application of machine learning, a subset of artificial intelligence that enables computers to learn from and make predictions or decisions based on data. The objective of this research work is to investigate the feasibility of a machine learning-based decision-support system that can automatically evaluate the quality of perfusion during colorectal surgery. For this purpose, surgical videos were collected and preprocessed to reduce the impact of environmental factors. Randomly chosen regions of interest were tracked in the video frames to extract fluorescence intensity-time curves. Unsupervised machine learning methods including K-means and self-organizing map were developed to obtain curve patterns. This provides a starting point to perform risk labelling on each pattern and feed a neural network that may classify a bowel area as adequately or inadequately perfused. Such a system could facilitate efficient and objective decision-making in the operating room.

Downloads

Actions (login required)

Modifica documento