30 August 2021 in IEEE Transactions on Engineering Management
Lung diseases are a tremendous challenge to the health and life of people globally, accounting for 5 out of 30 most common causes of death. Early diagnosis is crucial to help in faster recovery and improve long-term survival rates. Deep learning techniques offer a great promise for automated, fast, and reliable detection of lung diseases from medical images. Specifically, convolutional neural networks have accomplished encouraging results in disease detection. In spite of that, the performance of such supervised models depends heavily on the availability of large labeled data, the collection of which is an expensive and tedious task, specially for a novel disease. Therefore, in this article, we propose a deep unsupervised framework to classify lung diseases from chest CT and X-ray images. Our framework introduces multiple-layer generative adversarial networks called Lung-GANs that learn interpretable representations of lung disease images using only unlabeled data. We use the lung features learned by the model to train a support vector machine and a stacking classifier. We demonstrate through experiments that the proposed method outperforms the current state-of-the-art unsupervised models in lung disease classification. Our model obtained an accuracy of 94%–99.5% on all the six large-scale publicly available lung disease datasets used in this study. Hence, the proposed framework will simplify lung disease detection by reducing the time for diagnosis and increasing the convenience of diagnostics.
14 September 2021 in IEEE Transactions on Engineering Management
Recently computer-aided diagnosis methods have been widely adopted to aid doctors in disease diagnosis making their decisions more reliable and error-free. Electrocardiogram (ECG) is the most commonly used, noninvasive diagnostic tool for investigating various cardiovascular diseases. In real life, patients suffer from more than one heart disease at a time. So any practical automated heart disease diagnosis system should identify multiple heart diseases present in a single ECG signal. In this article, we propose a novel deep learning-based method for the multilabel classification of ECG signals. The proposed method can accurately identify up to two labels of an ECG signal pertaining to eight rhythm or morphological abnormalities of the heart and also the normal heart condition. Also, the black-box nature of deep learning models prevents them from being applied to high-risk decisions like the automated heart disease diagnosis. So in this article, we also establish an explainable artificial intelligence (XAI) framework for ECG classification using class activation maps obtained from the Grad-CAM technique. In the proposed method, we train a convolutional neural network (CNN) with constructed ECG matrices. With the experiments conducted, we establish that training the CNN by taking only one label for each ECG signal data point is enough for the network to learn the features of an ECG point with multilabel information in it (multiple heart diseases at the same time). During classification, we apply thresholding on the output probabilities of the softmax layer of our CNN, to obtain the multilabel classification of ECG signals.We trained the model with 6311 ECG records and tested the model with 280 ECG records. During testing, the model achieved a subset accuracy of 96.2% and a hamming loss of 0.037 and a precision of 0.986 and a recall of 0.949 and an F1-score of 0.967. Considering the fact that the model has performed very well in all the metrics of multilabel classification, the model can be directly used as a practical tool for automated heart disease diagnosis.
13 January 2023 in Engineering Applications of Artificial Intelligence
20 March 2025 in Computers in Biology and Medicine
12 May 2022 in Computational Intelligence