Computer-aided diagnosis with a convolutional neural network algorithm for automated detection of urinary tract stones on plain X-ray

Abstract

Background: Recent increased use of medical images induces further burden of their interpretation for physicians. A plain X-ray is a low-cost examination that has low-dose radiation exposure and high availability, although diagnosing urolithiasis using this method is not always easy. Since the advent of a convolutional neural network via deep learning in the 2000s, computer-aided diagnosis (CAD) has had a great impact on automatic image analysis in the urological field. The objective of our study was to develop a CAD system with deep learning architecture to detect urinary tract stones on a plain X-ray and to evaluate the model's accuracy.

Methods: We collected plain X-ray images of 1017 patients with a radio-opaque upper urinary tract stone. X-ray images (n = 827 and 190) were used as the training and test data, respectively. We used a 17-layer Residual Network as a convolutional neural network architecture for patch-wise training. The training data were repeatedly used until the best model accuracy was achieved within 300 runs. The F score, which is a harmonic mean of the sensitivity and positive predictive value (PPV) and represents the balance of the accuracy, was measured to evaluate the model's accuracy.

Results: Using deep learning, we developed a CAD model that needed 110 ms to provide an answer for each X-ray image. The best F score was 0.752, and the sensitivity and PPV were 0.872 and 0.662, respectively. When limited to a proximal ureter stone, the sensitivity and PPV were 0.925 and 0.876, respectively, and they were the lowest at mid-ureter.

Conclusion: CAD of a plain X-ray may be a promising method to detect radio-opaque urinary tract stones with satisfactory sensitivity although the PPV could still be improved. The CAD model detects urinary tract stones quickly and automatically and has the potential to become a helpful screening modality especially for primary care physicians for diagnosing urolithiasis. Further study using a higher volume of data would improve the diagnostic performance of CAD models to detect urinary tract stones on a plain X-ray.

Keywords: Artificial intelligence; Convolutional neural network; Deep learning; Urinary tract stone; Urolithiasis; X-ray.

Fig. 1

Flowchart of the inclusion and exclusion criteria and study outline. Eight hundred and twenty-seven X-ray images were used for training and 190 X-ray images were used for evaluating the model’s accuracy

Fig. 2

Labeling stone lesions and image division into patches. a Resized plain X-ray image of a patient with a left ureteral stone. b Labeling of stone lesions by urologists. A blue area in the image is a label showing the correct location of the stone lesion. c Random cropping and creating patches. Patches of 166 × 166 pixels were randomly cropped from a plain X-ray image and divided into two groups: patches including or not including a stone lesion

Fig. 3

ResNet architecture. The patches were input and convoluted as they passed through each layer. Each box indicates the number (n) and size (length (l) × width (w) = pixels) of images in each layer. The computer’s prediction of whether an input patch was included was output and each loss was calculated if the output was not concordant with the input. The parameters were optimized using the back propagation method, in which each loss was supposed to be minimized

Fig. 4

Preparation to evaluate the model’s accuracy. a Heat map representing the possibility of a stone lesion by color between light red at 100% and dark green at 0%. b Bounding boxes were automatically created to enclose three pixels outside of the heat maps

Fig. 5

Visualization of four representative cases. a A case with multiple calculi including a mid-ureteral stone. b A case in which a calculus was able to be distinguished from pelvic phleboliths. c A case with residual barium in the colon. d A case with multiple calculi and an artificial joint

Fig. 6

The models’ diagnostic performance that was created for each weight of loss for overlooking. This line graph indicates that the sensitivity was increased and that the PPV and F score were decreased as the weight of loss for overlooking was increased