Towards early bowel cancer detection: A data driven dynamic method of lesions characterization using a robotic capsule

Recent advances in miniaturised, dynamically actuated micro-robots have opened new possibilities for non-visual, in-situ disease diagnosis. This study introduces a method for early bowel cancer detection using a self-propelled robotic capsule autonomously navigating the bowel and identifying maligna...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2025 30th International Conference on Automation and Computing (ICAC) S. 1 - 6
Hauptverfasser: Afebu, Kenneth Omokhagbo, Liu, Yang, Papatheou, Evangelos, Prasad, Shyam
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 27.08.2025
Schlagworte:
Accuracy
Artificial networks
Biomechanical evaluation
Biomechanics
Bowel cancer
Cancer detection
Capsule robot
Computational modeling
Data models
Early diagnosis
Lesions
Support vector machines
Uncertainty
Vibrations
Visualization
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Recent advances in miniaturised, dynamically actuated micro-robots have opened new possibilities for non-visual, in-situ disease diagnosis. This study introduces a method for early bowel cancer detection using a self-propelled robotic capsule autonomously navigating the bowel and identifying malignant lesions based on localised variations in tissue stiffness. The approach leverages the sensitivity of the capsule's dynamic responses to biomechanical changes in the surrounding tissue. A dual-phase machine learning framework is proposed. In the first phase, the capsule's displacement signals are processed into features which are then used to predict tissue stiffness (Young's modulus, E) using multilayer perceptron (MLP), support vector regression (SVR), and Gaussian process regression (GPR). In the second phase, a Gaussian mixture model (GMM) is used to cluster the predicted E-values into discrete stiffness categories. Both accuracy and uncertainty Oriented metrics were used to evaluate the performances of the models. MLP offered the most robust regression on simulated data, while GPR outperformed it on experimental data. The GMM achieved over 89% clustering accuracy across both datasets. This study highlights a promising route for dynamic, non-visual bowel cancer detection and sets the stage for reliable deep tissue biopsy, while also addressing the current drawbacks of visual-based bowel cancer diagnostic methods.
DOI:10.1109/ICAC65379.2025.11196111