The method for breath detection using gas sensor array is gaining popularity. It was found that acetone can be used as breath marker in diabetic patients. In this paper, seven metal oxide gas sensors were used to collect acetone and ethanol gas which are used to simulate the exhaled breath of diabetic patients to obtain multidimensional response data. The kernel principal component analysis (KPCA) algorithm is used to extract characteristics from the data collected by the sensor array. The kind of gases is qualitatively identified using the AdaBoost algorithm, and the Grid Search method is used to automatically optimize the parameters of AdaBoost algorithm. Quantitative analysis of gas concentration is performed using multivariate relevance vector machines (MVRVM) and it is also trained using the gas sensor array drift dataset at different concentrations from the UCI database. The experimental results show that the accuracy of the algorithm in the qualitative identification of acetone and ethanol gas reaches 99.722%, and the root mean square errors for quantification of acetone and ethanol gases are 0.027 and 0.030, respectively. The algorithm is used for qualitative identification on the gas sensor array drift dataset at different concentrations with an accuracy of 94.55%, and the root mean square errors for quantification of acetone and ethanol gases are 11.59 and 8.72, respectively.

The application of artificial intelligence to point-of-care testing (POCT) disease detection has become a hot research field, in which breath detection, which detects the patient’s exhaled VOCs, combined with sensor arrays of convolutional neural network (CNN) algorithms as a new lung cancer detection is attracting more researchers’ attention. However, the low accuracy, high-complexity computation and large number of parameters make the CNN algorithms difficult to transplant to the embedded system of POCT devices. A lightweight neural network (LTNet) in this work is proposed to deal with this problem, and meanwhile, achieve high-precision classification of acetone and ethanol gases, which are respiratory markers for lung cancer patients. Compared to currently popular lightweight CNN models, such as EfficientNet, LTNet has fewer parameters (32 K) and its training weight size is only 0.155 MB. LTNet achieved an overall classification accuracy of 99.06% and 99.14% in the own mixed gas dataset and the University of California (UCI) dataset, which are both higher than the scores of the six existing models, and it also offers the shortest training (844.38 s and 584.67 s) and inference times (23 s and 14 s) in the same validation sets. Compared to the existing CNN models, LTNet is more suitable for resource-limited POCT devices.

Sensor array is the core of electronic nose (E-nose) system, while pattern recognition algorithm is an important component of the system. So, the sensor array optimization and the algorithm research cannot emphasize the importance too much. Volatile organic compounds (VOCs) mixture are ubiquitous in environmental monitoring, diseases screening, industrial emissions, and natural environments. Due to the flammability and potential health impacts, rapid and accurate detection of VOCs are crucial. Currently, E-nose systems for gas detection build separate models and random number of sensors for classification and concentration prediction, which lead to low efficiency and increased complexity. To address these issues, this study proposed a novel multi-task learning (MTL) framework combined with gated recurrent unit (GRU) and feature enhancement module (MTL-GRUA), performing both gas classification and concentration prediction tasks simultaneously. The model achieved over 99% accuracy in classification tasks for acetone, ethanol, and their mixture, and demonstrated high precision in concentration prediction tasks. By optimizing the hyperparameters of the network structure of MTL-GRUA by a particle swarm optimization (PSO) algorithm and using first 100s data instead of stable phase data, the detection efficiency was increased by 63%, achieving rapid detection. Furthermore, the experiment verified that even with an optimization in the number of sensors from 8 to 3, acceptable detection results could still be obtained. The application of multi-task learning proposed in this work provide a more efficient method for gas detection task.

The application of artificial intelligence to point-of-care testing (POCT) disease detection has become a hot research field, in which breath detection which detects the patient's exhaled VOCs combined with sensor arrays of Convolutional Neural Network (CNN) algorithms as a new lung cancer detection is attracting more researchers' attention. However, due to low accuracy, high complexity computation and large number of parameters make the CNN algorithms difficult transplant to the embedded system of POCT devices. A lightweight neural network (LTNet) in this work is proposed to deal with this problem, and meanwhile, achieve high-precision classification of acetone and ethanol gases, which are respiratory markers for lung cancer patients. Compared to currently popular lightweight CNN models such as EfficientNet, LTNet has fewer parameters (32K) and its training weight size is only 0.155 MB. LTNet achieved an overall classification accuracy of 99.06% and 99.14% in the own mixed gas dataset and the University of California (UCI) dataset, which are both higher than that of the five existing models’, and it also offers the shortest training (844.38s and 584.67s) and inference time (23s and 14s) in the same validation set. Compared to the existing CNN models, LTNet is more suitable for resource-limited POCT devices.