Deep learning with non-medical training used for chest pathology identification

By Yaniv Bar, Idit Diamant, Lior Wolf, Hayit Greenspan. 2015

325 citations recorded by the source corpus.

In this work, we examine the strength of deep learning approaches for pathology detection in chest radiograph data. Convolutional neural networks (CNN) deep architecture classification approaches have gained popularity due to their ability to learn mid and high level image representations. We explore the ability of a CNN to identify different types of pathologies in chest x-ray images. Moreover, since very large training sets are generally not available in the medical domain, we explore the feasibility of using a deep learning approach based on non-medical learning. We tested our algorithm on a dataset of 93 images. We use a CNN that was trained with ImageNet, a well-known large scale nonmedical image database. The best performance was achieved using a combination of features extracted from the CNN and a set of low-level features. We obtained an area under curve (AUC) of 0.93 for Right Pleural Effusion detection, 0.89 for Enlarged heart detection and 0.79 for classification between healthy and abnormal chest x-ray, where all pathologies are combined into one large class. This is a first-of-its-kind experiment that shows that deep learning with large scale non-medical image databases may be sufficient for general medical image recognition tasks.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

A survey on deep learning in medical image analysis (2017, methodology): “Application Remarks Lo et al. (1995) Nodule detection Classiﬁes candidates from small patches with two-layer CNN, each with 12 5 × 5 ﬁlters Anavi et al. (2015) Image retrieval Combines classical features with those from pre-trained CNN for image retrieval using SVM Bar et al. (2015) Pathology detect”
A Deep Feature Fusion Methodology for Breast Cancer Diagnosis Demonstrated on Three Imaging Modality Datasets (2017, unknown): “Other works have shown that the final layer of AlexNet is significantly inferior to earlier layers for the task of medical image analysis.(6,7) Our method employs a more advanced feature extraction technique by hierarchically integrating multiple layers from VGGNet in order to incorporate both low- ”
Using Non-invertible Data Transformations to Build Adversary-Resistant Deep Neural Networks (2016, background): “Recent interest has been to integrate it into critical fields like medical imaging [1], [37] and self-driving cars [13], [8].”
Deep Learning Techniques for Medical Image Segmentation: Achievements and Challenges (2019, methodology): “In another experiment done by Bar et al [4], a transfer learning approach is taken into account and low-level features are borrowed from a pre-trained model on Imagenet.”
CNNs for automatic glaucoma assessment using fundus images: an extensive validation (2019, methodology): “[19] pre-trained CNNs that were used as a feature extractor for chest pathology identification.”
Automatic Detection and Classification of Colorectal Polyps by Transferring Low-Level CNN Features From Nonmedical Domain (2017, background): “The transfer of CNN features can be selective, from different layers of the CNN [23].”
Deep learning in bioinformatics (2016, methodology): “[132] performed transfer learning using natural images from the ImageNet database [189] as pre-training data and fine-tuned with chest X-ray images to identify chest pathologies and to classify healthy and abnormal images.”
High-Throughput Classification of Radiographs Using Deep Convolutional Neural Networks (2016, background): “Other recent uses of recurrent neural networks to caption chest radiographs have assumed a dataset with correct and pre-defined metadata, which we have not found in our clinical database [15], or have used orders of magnitude fewer images than our work [8].”
Computer-aided detection in chest radiography based on artificial intelligence: a survey (2018, methodology): “[75] Custom Right pleural effusion, cardiomegaly, health, and abnormal disease AUC 93%, 89%, and 79%, respectively”
Deep Learning and Medical Diagnosis: A Review of Literature (2018, unknown): “[30] CNN X-ray Chest pathology identification; the area under the curve for heart detection was 0.”
Deep learning in medical imaging and radiation therapy. (2018, methodology): “A number of studies 173,181,192,196,306 – 308 extracted the outputs of the fully connected layers of a DL network that has been pretrained ImageNet, and used those features as input to SVMs to build classification models, which suggests that a network pretrained on natural images is useful for…”
How much data is needed to train a medical image deep learning system to achieve necessary high accuracy (2015, background): “This key question was not explored systematically in the recent medical image deep learning publications (Anavi et al. (2015), Havaei et al. (2015),Bar et al. (2015), Roth et al. (2015a), Yan et al. (2015), Roth et al. (2015c), Shin et al. (2015), Wang et al. (2015), Xu et al. (2015), Srivastava et…”
Convolutional neural networks based transfer learning for diabetic retinopathy fundus image classification (2017, methodology): “In paper [6], a similar research is also conducted.”