Fig. 2
From: Ensemble deep learning architectures for detecting pulmonary tuberculosis in chest X-rays
Overview of the proposed architectures used for feature extraction and classification for automated pulmonary tuberculosis detection. (a) Pre-processing: Chest X-rays undergo enhancement with histogram equalisation and lung field cropping to improve image quality; (b) An autoencoder is trained on chest X-ray images, with latent variables from the bottleneck layer for subsequent classification; (c) Classifier: classification is performed; (d) Multiscale Convolutional Neural Network: End-to-end classification is conducted, with features extracted from the last convolutional layer for ensemble learning.
