Vibroacoustic detection of inclusions in an elastomeric tissue phantom using a multilayer perceptron classifier: A...

Area of Science:

• Biomedical Engineering
• Surgical Technology
• Machine Learning in Medicine

Background:

• Accurate tumor boundary detection is crucial for oncologic surgery success.
• Traditional palpation is not feasible in Minimally Invasive Surgery (MIS).
• Need for non-invasive methods to assess tissue mechanical properties in MIS.

Purpose of the Study:

• To investigate the feasibility of detecting stiffer inclusions (simulating tumors) using vibroacoustic signatures.
• To develop a real-time detection method for simulated tumors in tissue phantoms.
• To evaluate a wavelet-based multilayer perceptron neural network for inclusion classification.

Main Methods:

• Utilized a static measurement probe and a micro-electro-mechanical system (MEMS) sensor.
• Acquired short-time (1s) vibroacoustic signals from simulated tissue phantoms with stiffer inclusions.
• Employed a supervised learning approach with a wavelet-based multilayer perceptron (MLP) for binary classification.

Main Results:

• The MLP classifier achieved a 75% F1 score for inclusion detection.
• The model demonstrated 77.8% accuracy on previously unseen test data.
• Performance was comparable to Support Vector Machine (SVM) classifiers.

Conclusions:

• The study supports the hypothesis that altered vibroacoustic signatures can indicate stiffer inclusions.
• The developed method shows promise as a proof-of-concept for tumor detection in MIS.
• Future work includes system enhancement and validation on biological tissues.