Discrimination of vocal folds lesions by multiclass classification using autofluorescence spectroscopy: An ex vivo study.

BACKGROUND: Autofluorescence spectroscopy is effective for noninvasive detection but underutilized in tissue with various pathology analyses. This study evaluates whether AFS can be used to discriminate between different types of laryngeal lesions in view of assisting in vocal fold surgery and preoperative investigations. METHODS: A total of 1308 spectra were recorded from 29 vocal fold samples obtained from 23 patients. Multiclass analysis was performed on the spectral data, categorizing lesions into normal, benign, dysplastic, or carcinoma. RESULTS: Through an appropriate selection of spectral components and a cascading classification approach based on artificial neural networks, a classification rate of 97% was achieved for each lesion class, compared to 52% using autofluorescence intensity. CONCLUSIONS: The ex vivo study demonstrates the effectiveness of AFS combined with multivariate analysis for accurate classification of vocal fold lesions. Comprehensive analysis of spectral data significantly improves classification accuracy, such as distinguishing malignant from precancerous or benign lesions.

CMUT-Based Sensor for Acoustic Emission Application: Experimental and Theoretical Contributions to Sensitivity Optimization.

The paper deals with a capacitive micromachined ultrasonic transducer (CMUT)-based sensor dedicated to the detection of acoustic emissions from damaged structures. This work aims to explore different ways to improve the signal-to-noise ratio and the sensitivity of such sensors focusing on the design and packaging of the sensor, electrical connections, signal processing, coupling conditions, design of the elementary cells and operating conditions. In the first part, the CMUT-R100 sensor prototype is presented and electromechanically characterized. It is mainly composed of a CMUT-chip manufactured using the MUMPS process, including 40 circular 100 µm radius cells and covering a frequency band from 310 kHz to 420 kHz, and work on the packaging, electrical connections and signal processing allowed the signal-to-noise ratio to be increased from 17 dB to 37 dB. In the second part, the sensitivity of the sensor is studied by considering two contributions: the acoustic-mechanical one is dependent on the coupling conditions of the layered sensor structure and the mechanical-electrical one is dependent on the conversion of the mechanical vibration to electrical charges. The acoustic-mechanical sensitivity is experimentally and numerically addressed highlighting the care to be taken in implementation of the silicon chip in the brass housing. Insertion losses of about 50% are experimentally observed on an acoustic test between unpackaged and packaged silicon chip configurations. The mechanical-electrical sensitivity is analytically described leading to a closed-form amplitude of the detected signal under dynamic excitation. Thus, the influence of geometrical parameters, material properties and operating conditions on sensitivity enhancement is clearly established: such as smaller electrostatic air gap, and larger thickness, Young's modulus and DC bias voltage.

Joint Prediction of Continuous and Discrete States in Time-Series Based on Belief Functions.

Forecasting the future states of a complex system is a complicated challenge that is encountered in many industrial applications covered in the community of prognostics and health management. Practically, states can be either continuous or discrete: Continuous states generally represent the value of a signal while discrete states generally depict functioning modes reflecting the current degradation. For each case, specific techniques exist. In this paper, we propose an approach based on case-based reasoning that jointly estimates the future values of the continuous signal and the future discrete modes. The main characteristics of the proposed approach are the following: 1) It relies on the K-nearest neighbor algorithm based on belief function theory; 2) belief functions allow the user to represent his/her partial knowledge concerning the possible states in the training data set, particularly concerning transitions between functioning modes which are imprecisely known; and 3) two distinct strategies are proposed for state prediction, and the fusion of both strategies is also considered. Two real data sets were used in order to assess the performance in estimating future breakdown of a real system.