Interpreting chest X-rays via CNNs that exploit disease dependencies and uncertainty labels

Abstract

Chest radiography is one of the most common types of diagnostic radiology exams, which is critical for screening and diagnosis of many different thoracic diseases. Specialized algorithms have been developed to detect several specific pathologies such as lung nodule or lung cancer. However, accurately detecting the presence of multiple diseases from chest X-rays (CXRs) is still a challenging task. This paper presents a supervised multi-label classification framework based on deep convolutional neural networks (CNNs) for predicting the risk of 14 common thoracic diseases. We tackle this problem by training state-of-the-art CNNs that exploit dependencies among abnormality labels. We also propose to use the label smoothing technique for a better handling of uncertain samples, which occupy a significant portion of almost every CXR dataset. Our model is trained on over 200,000 CXRs of the recently released CheXpert dataset and achieves a mean area under the curve (AUC) of 0.940 in predicting 5 selected pathologies from the validation set. This is the highest AUC score yet reported to date. The proposed method is also evaluated on the independent test set of the CheXpert competition, which is composed of 500 CXR studies annotated by a panel of 5 experienced radiologists. The performance is on average better than 2.6 out of 3 other individual radiologists with a mean AUC of 0.930, which ranks first on the CheXpert leaderboard at the time of writing this paper.

Validation of expert system enhanced deep learning algorithm for automated screening for COVID-Pneumonia on chest X-rays

“…Pathology detection on chest X-rays is performed using DenseNet-201 [13] as a backbone network. The input CxR image is resized to 224 x 224 using the linear interpolation [25] technique and was initialized with pretrained weights of imagenet dataset using transfer learning approach similar to the technique used in [26]. The pathology detection module outputs the probability corresponding to each pathology which is taken into consideration by the rule based COVID-19 likelihood detector to give final prediction.…”

Recalibration of deep learning models for abnormality detection in smartphone-captured chest radiograph

“…We trained the Model-TRNS and Model-PHOT using 10 and 50 epochs, respectively, and after that the model was converged. For the Model-ORIG, the training dataset was augmented by a random transformation (rotating ±7 degrees, scaling ±2%, and shearing ±5 pixels) twice 38 . For the Model-RECA, we augmented the training dataset using our augmentation functions with two sets of hyperparameters, which were determined by complex wavelet structural similarity method 33 and the Bhattacharyya distance of image histogram, respectively.…”

Corona Virus Detection and Classification with radiograph images using RNN

“…Meanwhile, researchers are working on a unified deep learning system for accurately de-tecting several common thoracic diseases [7]. Reverse tran-scription poly-merase chain reaction [8] confirms the diag-nosis of COVID-19. The importance of chest radiography (CXR) is still a hot topic of debate.…”

“…This can lead to CXR results be-ing misinterpreted, rendering comparisons between exami-nations and study studies difficult [8,6]. We recommend ter-minology for consistent CXR reporting and severity assess-ment of individuals under investigation for COVID-19, pa-tients with a verified diagnosis of COVID-19, and patients who may have radiographic symptoms of COVID-19, in or-der to satisfy this need for accuracy.When the diagnosis of COVID-19 is not suspected clinically, re-sults characteristic or indicative of COVID-19 are found [8]. The most common imaging test in the country is chest ra-diography, which is essential for screening, diagnosing, and treating a variety of life-threatening diseases [9].…”

“…The most popular image analysis requested is chest X-rays (CXRs).CXR Com-puterAided Diagnosis (CAD) has gotten a lot of attention in the scientific community, both be-fore and after the suc-cess of deep learning [10]. A recent effort has been made to build a new generation of CAD systems for the identification and visualization of common thoracic diseases from CXR images using advances in machine learning, especially deep learning [11,10,8].The area of time series forecasting known as spatial-temporal time series is concerned with variables that shift over time and space. Disease forecasting, such as COVID-19, would benefit from spatial-temporalforecasting because it analyses patterns and provides a reliable predic-tor for decisionmakers all over the world, ability to make the right decisions at the right time [12].…”