Learning to Recognize Chest-Xray Images Faster and More Efficiently Based on Multi-Kernel Depthwise Convolution

Abstract

The development of convolutional neural networks has promoted the progress of computeraided diagnostic systems. Details in medical image, such as the texture and tissue structure, are crucial features for diagnosis. Therefore, large input images combined with deep convolution neural networks are adopted to boost the performance in recent research of chest X-ray diagnosis. Meanwhile, due to the variable sizes of thoracic diseases, many researchers have worked to introduce additional module to capture multi-scale feature of images in CNN. However, these efforts hardly consider the computational costs of large inputs and introduced additional modules. This paper aims to automatically diagnose diseases on chest X-rays images quickly and effectively. We propose the multi-kernel depthwise convolution(MD-Conv) which contains depthwise convolution kernels with different filter sizes in one depthwise convolution layer. MD-Conv has high calculation efficiency and few parameters. Because its ability to learn multi-scale feature based on the multi-size kernels, it is appropriate for medical images diagnosis tasks in which abnormalities varied in sizes. In addition, larger depthwise convolution kernels are adopted in MD-Conv to obtain a larger receptive field efficiently, which can ensure sufficient receptive field for high resolution inputs. MD-Conv can be easily applied in modern lightweight networks to replace the normal depthwise convolution layer. We conduct experiments on the Chest X-ray 14 Dataset, which is the largest available chest x-ray dataset, and obtain competitive results. We also evaluate the MD-Conv on the new released dataset for pediatric pneumonia diagnosis. We obtain a better performance of 98.3% AUC than original paper (96.8%) for recognize pneumonia versus normal. Meanwhile we compare the FLOPs and Params of different models to show their efficiency for chest X-rays recognition. INDEX TERMS Chest x-ray recognition, lightweight networks, multi-kernels depthwise convolution.

Efficient Pediatric Pneumonia Diagnosis Using Depthwise Separable Convolutions

“…Recently, Hu et al [43] have attempted to develop a faster and more efficient pneumonia detection model based on Multi-kernel Depthwise Convolution (MD-Conv). In order to achieve this, MobileNetV2 [16] was modified to incorporate MD-Conv in its design.…”

“…Hu et al [43] argue that due to the use of multi-size kernels, their model is better suited to learning multi-scale features present in CXR images. In addition, Hu et al [43] also argue that larger kernel size (i.e. use of 5 ×5 kernel) will ensure sufficient receptive field for high resolution inputs.…”

“…However, it is worth noting that even though the model's performance is acceptable, it is still relatively inferior to that reported for the models of Siddiqi [21] and Rajaraman et al [23], especially the recall values which are crucial in correct disease diagnosis. In their paper [43], Hu et al have emphasized that their model is more efficient than ResNet50 [11] and DenseNet121 [15] in terms of the total number of parameters as well as FLOPs (Floating Point Operations per Second). However, it is also important to note that their model [43] has higher FLOPs when compared with other lightweight models like the original MobileNetV2 [16].…”

“…PneumoniaNet is proven to be more efficient than all other known models for this task. These even include other lightweight models like the customized Mobile-NetV2 and also the recently developed model [43] based on multi-kernel depthwise convolution. 3.…”

An Overview of Deep Learning Approaches in Chest Radiograph

“…X-rays are electromagnetic radiations that penetrate human body producing an image of the internal structures. It distinguishes objects based on its density; the four basic densities found on a radiograph are bone (appears white), soft tissues (appears white to gray), fat (appears gray) and gas (appears black) [7]. The penetrations of X-ray through these densities depend on the exposure and power of the beam.…”