Opto-Acoustical Feedback System for Smart Laser Surgery

Characterizing acoustic shock waves (ASWs) for tissue differentiation is of vital interest for developing an opto-acoustical feedback system during laser surgery. This is particularly true if the laser system is not controlled by an in situ and real-time feedback control system that is not only able to differentiate specific types of human tissues but can automatically stop.
The research presented in this thesis focused on developing an opto-acoustical feedback sensor and on designing an efficient optical method for detecting ASWs for tissue differentiation. The method has advantages to extract physical properties of tissues based on measured ASW features. Some of them are the shock wave rising- and falling-time and the acoustic amplitude-spectrum extracted from the measured emitted shock/acoustic signals.
The work included novel aspects, such as measuring the ASWs generated using advanced custom-made optical technologies, characterizing the measurements by looking at the amplitude frequency band which provides the least average classification error. We used principal component analysis (PCA) combined with advanced signal processing to classify tissue types. Since PCA allows us to keep essential features (variance) of data by reducing their dimensionality, we expected to shorten the time needed to train a classifier and to avoid overfitting (less accuracy when classifying the validation data as compared to the training data). In this case, the first principal components (PCs) are dominated by the high-variance variables and mostly represent variance— i.e. the acoustic amplitude-spectrum of tissues — of each data.
The approach evolved over three steps. In the first step, four tissue types were initially classified during laser ablation by measuring the ASWs generated with a conventional air-coupled transducer and by processing the information using the Mahalanobis distance method. Later, five tissue types were classified using an artificial neural network (ANN), and quadratic or Gaussian support-vector machine methods (SVM) combined with principal component analysis (PCA) during classification experiments. It was possible to differentiate hard tissue from soft tissue types, but distinguishing between soft tissues remained a challenge.
In the second step, classification errors between soft tissue types were reduced. This was made possible by building a Mach–Zehnder interferometer that acts as a high-frequency microphone to provide accurate measurements of ASWs. PCA and the Mahalanobis distance method were used to differentiate among scores of measured ASWs. The results of this study demonstrated a promising technique for differentiating tissues during laser osteotomy.
The final step was to design and build a fiber-coupled Fabry-Pérot etalon sensor as an alternative compact optical sensor for measuring ASWs. The miniaturized etalon cavity was built to fit into a 5 mm diameter endoscope for minimally invasive smart laser osteotome. The collected data were subsequently investigated by looking at the amplitude frequency band to find the minimum classification error. Tissue classification was performed using PCA combined with the artificial neural network. Based on these results, we argue that this method can be used in endoscopic applications for tissue classification.
This study was part of the Minimally Invasive Robot-Assisted Computer-guided LaserosteotomE (MIRACLE) project. Thus, measuring ASWs using a custom-made fiber-optical-based acoustic sensor for tissue classification was the main goal of this research.