Deep-learning framework to detect lung abnormality – A study with chest X-Ray and lung CT scan images

• Published in: Pattern recognition letters Vol. 129; pp. 271 - 278
• Main Authors: Bhandary, Abhir, Prabhu, G. Ananth, Rajinikanth, V., Thanaraj, K. Palani, Satapathy, Suresh Chandra, Robbins, David E., Shasky, Charles, Zhang, Yu-Dong, Tavares, João Manuel R.S., Raja, N. Sri Madhava
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.01.2020; Elsevier Science Ltd
• Subjects: Abnormalities; Chest; Classification; Computed tomography; Deep learning; Image classification; Lung cancer; Medical imaging; Pneumonia; Principal components analysis; Support vector machines
• ISSN: 0167-8655; 1872-7344
• DOI: 10.1016/j.patrec.2019.11.013

•This work proposes a Modified AlexNet (MAN) deep-learning framework to evaluate the lung abnormality.•This work introduces a threshold filter to remove the artifacts from the Lung CT images.•This work introduces an Ensemble-Feature-Technique (EFT) by integrating the deep-features and the handcrafted features.•Serial fusion and PCA based selection is implemented in EFT to chose principal feature set.•Experimental results demonstrate superior performance of MAN in comparison with other existing state of the art methods. Lung abnormalities are highly risky conditions in humans. The early diagnosis of lung abnormalities is essential to reduce the risk by enabling quick and efficient treatment. This research work aims to propose a Deep-Learning (DL) framework to examine lung pneumonia and cancer. This work proposes two different DL techniques to assess the considered problem: (i) The initial DL method, named a modified AlexNet (MAN), is proposed to classify chest X-Ray images into normal and pneumonia class. In the MAN, the classification is implemented using with Support Vector Machine (SVM), and its performance is compared against Softmax. Further, its performance is validated with other pre-trained DL techniques, such as AlexNet, VGG16, VGG19 and ResNet50. (ii) The second DL work implements a fusion of handcrafted and learned features in the MAN to improve classification accuracy during lung cancer assessment. This work employs serial fusion and Principal Component Analysis (PCA) based features selection to enhance the feature vector. The performance of this DL frame work is tested using benchmark lung cancer CT images of LIDC-IDRI and classification accuracy (97.27%) is attained. [Display omitted]