Detection of pneumoperitoneum in the abdominal radiograph images using artificial neural networks

Abstract

Highlights The purpose of this study was to assess the diagnostic performance of artificial neural networks (ANNs) to detect pneumoperitoneum in abdominal radiographs for the first time. This approach applied a novel deep-learning algorithm, a simple ANN training process without employing CNN, and also used ResNet-50, for comparison. By applying ResNet-50 to abdominal radiographs, we obtained an area under the ROC curve (AUC) of 0.916 and an accuracy of 85.0 % with a sensitivity of 85.7 % and a predictive value of the negative tests (NPV) of 91.7 %. Compared with CNN, our novel approach used extremely small ANN structures and a simple ANN training process. The diagnostic performance of our approach, with a sensitivity of 88.6 % and NPV of 91.3 %, was compared decently with that of ResNet-50.

Detection of the location of pneumothorax in chest X-rays using small artificial neural networks and a simple training process

“…As the pixel resolution of the ANN input images increased from 20 × 20 pixels, its diagnostic performance for detecting the location of pneumothorax reached its highest value at 30 × 30 pixels and then degraded at higher resolutions. Therefore, rather than simply increasing the resolution of input images for ANNs, reducing the input resolution to an appropriate level could yield better results 16 , 18 .…”

“…To obtain a fully-connected small ANN structure with the optimal diagnostic performance, the input image resolution and the number of hidden layers in the ANN model started at 20 × 20 and 1, respectively, and were gradually increased 17 , 18 to compare the resulting performances with those of other ANN models. Therefore, the pixel resolution of the original chest X-ray images (512 × 512) was reduced to 20 × 20, 30 × 30, 40 × 40, 60 × 60, and 80 × 80 pixels.…”

“…A training session was terminated after 10 repetitions of the training cycle, during which the variable selection ratio of the training cycle steadily decreased from 15 to 1.5% 16 of the total number of variables in the ANN or CNN. During the training cycle, the total sum of the variable selection ratios was set to 900%, and 30 attempts were made to adjust the values of the selected variables by small random amounts 17 , 18 . After the training session was completed, the test set was applied to the ANN or CNN to obtain the test results and identify the diagnostic performance of the deep-learning model.…”

“…An author (J. S. Kim) developed this algorithm, which imitates the biological evolution of animals that adapt to a given environment according to the principle of the survival of the fittest 16 . The algorithm, a simple training process for ANNs, has been applied to predict the location of the glottis in laryngeal images 16 , to detect and classify intracranial haemorrhage on CT images 17 , and to detect pneumoperitoneum in abdominal radiograph images 18 . For data classification problems, there have also been studies using ANNs containing morphological neurons 19 , 20 .…”

“…The purpose of this study was to investigate the diagnostic performances of various fully-connected small ANN models for detecting the location of pneumothorax in chest X-rays using a simple training process (the Kim-Monte Carlo deep-learning algorithm). The dependence of diagnostic performance (in terms of detecting the location of pneumothorax) on the resolutions of the X-ray images input into the ANNs was also studied 16 , 18 . In addition, a common deep-learning method for image recognition, CNN, was also applied to the same dataset for comparison purposes.…”