Published papers

Objectives Renal cell carcinoma (RCC) is often found incidentally in asymptomatic individuals undergoing abdominal computed tomography (CT) examinations. The purpose of our study is to develop a deep learning-based algorithm for fully automated detection of small (≤4 cm) RCCs in contrast-enhanced CT images using a multicenter database and to evaluate its performance. Materials and Methods For the algorithmic detection of RCC, we retrospectively selected contrast-enhanced CT images of patients with histologically confirmed single RCC with a tumor diameter of 4 cm or less between January 2005 and May 2020 from 7 centers in the Japan Medical Image Database. A total of 453 patients from 6 centers were selected as dataset A, and 132 patients from 1 center were selected as dataset B. Dataset A was used for training and internal validation. Dataset B was used only for external validation. Nephrogenic phase images of multiphase CT or single-phase postcontrast CT images were used. Our algorithm consisted of 2-step segmentation models, kidney segmentation and tumor segmentation. For internal validation with dataset A, 10-fold cross-validation was applied. For external validation, the models trained with dataset A were tested on dataset B. The detection performance of the models was evaluated using accuracy, sensitivity, specificity, and the area under the curve (AUC). Results The mean ± SD diameters of RCCs in dataset A and dataset B were 2.67 ± 0.77 cm and 2.64 ± 0.78 cm, respectively. Our algorithm yielded an accuracy, sensitivity, and specificity of 88.3%, 84.3%, and 92.3%, respectively, with dataset A and 87.5%, 84.8%, and 90.2%, respectively, with dataset B. The AUC of the algorithm with dataset A and dataset B was 0.930 and 0.933, respectively. Conclusions The proposed deep learning-based algorithm achieved high accuracy, sensitivity, specificity, and AUC for the detection of small RCCs with both internal and external validations, suggesting that this algorithm could contribute to the early detection of small RCCs.

Purpose: This study aims to elucidate correlation between heart volume on computed tomography (CT) and various health checkup examination data in the general population. Furthermore, this study aims to examine the utility of a deep-learning segmentation tool in the data-driven analysis of CT big data. Materials and methods: Health checkup examination data and CT images acquired in 2013 and 2018 were retrospectively analyzed. We first quantified heart volume using a public deep-learning model, TotalSegmentator. The accuracy of segmentation was evaluated using Dice score on 30 randomly chosen images and annotation by a radiologist. Then, Spearman’s partial correlation was calculated for 58 numerical items, and the analysis of covariance was performed for 13 categorical items, adjusting for the effect of gender, medication, height, weight, abdominal circumference, and age. The variables found to be significant proceeded to longitudinal analysis. Results: In the dataset, 7993 records were eligible for cross-sectional analysis and 1306 individuals were eligible for longitudinal analysis. Pulse rate was most strongly inversely correlated with the heart volume (Spearman’s correlation coefficients ranging from – 0.29 to – 0.33). A 10 bpm increase in pulse rate was correlated with roughly a 0.5 percentage point decrease in the cardiothoracic ratio. Hemoglobin, hematocrit, total protein, albumin, and cholinesterase also showed weak inverse correlation. Five-year longitudinal analysis corroborated these findings. Conclusions: We found that pulse rate was the strongest covariate of the heart volume on CT, rather than other cardiovascular-related variables such as blood pressure. The study also demonstrated the feasibility and utility of the artificial intelligence-assisted data-driven research on CT big data.

Purpose: To evaluate the performance of a deep learning-based computer-aided detection (CAD) software for detecting pulmonary nodules, masses, and consolidation on chest radiographs (CRs) and to examine the effect of readers’ experience and data characteristics on the sensitivity and final diagnosis. Materials and methods: The CRs of 453 patients were retrospectively selected from two institutions. Among these CRs, 60 images with abnormal findings (pulmonary nodules, masses, and consolidation) and 140 without abnormal findings were randomly selected for sequential observer-performance testing. In the test, 12 readers (three radiologists, three pulmonologists, three non-pulmonology physicians, and three junior residents) interpreted 200 images with and without CAD, and the findings were compared. Weighted alternative free-response receiver operating characteristic (wAFROC) figure of merit (FOM) was used to analyze observer performance. The lesions that readers initially missed but CAD detected were stratified by anatomic location and degree of subtlety, and the adoption rate was calculated. Fisher’s exact test was used for comparison. Results: The mean wAFROC FOM score of the 12 readers significantly improved from 0.746 to 0.810 with software assistance (P = 0.007). In the reader group with < 6 years of experience, the mean FOM score significantly improved from 0.680 to 0.779 (P = 0.011), while that in the reader group with ≥ 6 years of experience increased from 0.811 to 0.841 (P = 0.12). The sensitivity of the CAD software and the adoption rate for the lesions with subtlety level 2 or 3 (obscure) lesions were significantly lower than for level 4 or 5 (distinct) lesions (50% vs. 93%, P < 0.001; and 55% vs. 74%, P = 0.04, respectively). Conclusion: CAD software use improved doctors’ performance in detecting nodules/masses and consolidation on CRs, particularly for non-expert doctors, by preventing doctors from missing distinct lesions rather than helping them to detect obscure lesions.

Ultrasonography is widely used for diagnosis of diseases in internal organs because it is nonradioactive, noninvasive, real-time, and inexpensive. In ultrasonography, a set of measurement markers is placed at two points to measure organs and tumors, then the position and size of the target finding are measured on this basis. Among the measurement targets of abdominal ultrasonography, renal cysts occur in 20–50% of the population regardless of age. Therefore, the frequency of measurement of renal cysts in ultrasound images is high, and the effect of automating measurement would be high as well. The aim of this study was to develop a deep learning model that can automatically detect renal cysts in ultrasound images and predict the appropriate position of a pair of salient anatomical landmarks to measure their size. The deep learning model adopted fine-tuned YOLOv5 for detection of renal cysts and fine-tuned UNet++ for prediction of saliency maps, representing the position of salient landmarks. Ultrasound images were input to YOLOv5, and images cropped inside the bounding box and detected from the input image by YOLOv5 were input to UNet++. For comparison with human performance, three sonographers manually placed salient landmarks on 100 unseen items of the test data. These salient landmark positions annotated by a board-certified radiologist were used as the ground truth. We then evaluated and compared the accuracy of the sonographers and the deep learning model. Their performances were evaluated using precision–recall metrics and the measurement error. The evaluation results show that the precision and recall of our deep learning model for detection of renal cysts are comparable to standard radiologists; the positions of the salient landmarks were predicted with an accuracy close to that of the radiologists, and in a shorter time.