A deep learning approach using MobileNetV2, GoogLeNet, fMRMR feature selection, and SVM achieved 95.80% accuracy in classifying medical devices from the MedDev5 dataset.
A deep learning approach using CNN feature extraction and SVM classification achieved 95.8% accuracy in classifying medical device images, demonstrating the potential of AI in clinical engineering.
This study introduces a deep learning approach for classifying medical devices to help improve the accuracy of inventory information in health care facilities. For this purpose, we constructed and made publicly available the MedDev5 dataset, which includes images of medical devices such as anesthesia machines, ventilators, ECGs, and patient and fetal monitors. For experimental results, the features of MedDev5 images were first extracted using MobileNetV2 and GoogLeNet CNN models. These features were then concatenated to achieve high classification performance. Following this, to eliminate redundant and unnecessary features and select the most informative features, the fMRMR feature selection method was used. Finally, medical device image classification was performed using support vector machines (SVM). The experiments showed that the classification of the medical devices in the MedDev5 dataset was achieved with a high accuracy of 95.80% using 500 selected features. This result indicates that we may frequently encounter artificial intelligence applications in clinical engineering in the near future.
Muzoğlu et al. (Fri,) conducted a other in Medical device classification. Deep learning approach (MobileNetV2, GoogLeNet, fMRMR, SVM) was evaluated on Medical device image classification accuracy. A deep learning approach using MobileNetV2, GoogLeNet, fMRMR feature selection, and SVM achieved 95.80% accuracy in classifying medical devices from the MedDev5 dataset.