Clinical Practice Guidelines for Hepatocellular Carcinoma: The Japan Society of Hepatology 2021 version (5th JSH-HCC Guidelines).

The fifth version of the Clinical Practice Guidelines for Hepatocellular Carcinoma was revised by the Japan Society of Hepatology, according to the methodology of evidence-based medicine and partly to the Grading of Recommendations Assessment, Development and Evaluation system, which was published in October 2021 in Japanese. In addition to surveillance-diagnostic and treatment algorithms, a new algorithm for systemic therapy has been created, as multiple drugs for hepatocellular carcinoma can be currently selected. Here, new or revised algorithms and evidence on which the recommendations are based are described.

Predictors of rituximab effect on modified Rodnan skin score in systemic sclerosis: a machine-learning analysis of the DesiReS trial.

OBJECTIVES: The double-blind, parallel-group comparison, investigators initiated phase II clinical trial of IDEC-C2B8 (Rituximab) in patients with Systemic sclerosis (DesiReS) trial showed that rituximab is effective in treating skin sclerosis in SSc. However, which patient groups are likely to benefit from rituximab is unknown. METHODS: We performed post-hoc analysis of prospective data from 54 patients who received rituximab or placebo in the DesiReS trial. Twenty-seven baseline factors were used to investigate subpopulations with different magnitudes of rituximab effect on modified Rodnan skin score (mRSS) change at 24 weeks. Based on a machine-learning algorithm called the causal tree, we explored the combination of predictors needed to identify subpopulations that would respond to rituximab and have good treatment outcomes. RESULTS: Three factors were identified as branches of the decision tree: peripheral blood CD19-positive cell counts', 'mRSS', and 'serum surfactant protein D (SP-D) levels'. It was only in the subpopulation of patients with CD19-positive cell counts of <57/µl that rituximab did not show a significant improvement in mRSS vs placebo. In the subpopulation of patients with CD19-positive cell counts of ≥57/µl and mRSS ≥ 17, mRSS was most improved with rituximab [difference -17.06 (95% CI: -24.22, -9.89)]. The second greatest improvement in mRSS with rituximab was in the subpopulation with CD19-positive cell counts of ≥57/µl, mRSS < 17, and serum SP-D levels of ≥151 ng/ml [difference -10.35 (95% CI: -14.77, -5.93)]. CONCLUSION: SSc patients who have high CD19-positive cell counts and high mRSS are expected to have greater improvement in mRSS with rituximab. When the patients with high CD19-positive cell counts show low mRSS, serum SP-D levels may modify the treatment effect. TRIAL REGISTRATION: ClinicalTrials.gov, https://clinicaltrials.gov, NCT04274257 and UMIN-CTR; https://center6.umin.ac.jp, UMIN000030139.

Automatic detection of actionable radiology reports using bidirectional encoder representations from transformers.

BACKGROUND: It is essential for radiologists to communicate actionable findings to the referring clinicians reliably. Natural language processing (NLP) has been shown to help identify free-text radiology reports including actionable findings. However, the application of recent deep learning techniques to radiology reports, which can improve the detection performance, has not been thoroughly examined. Moreover, free-text that clinicians input in the ordering form (order information) has seldom been used to identify actionable reports. This study aims to evaluate the benefits of two new approaches: (1) bidirectional encoder representations from transformers (BERT), a recent deep learning architecture in NLP, and (2) using order information in addition to radiology reports. METHODS: We performed a binary classification to distinguish actionable reports (i.e., radiology reports tagged as actionable in actual radiological practice) from non-actionable ones (those without an actionable tag). 90,923 Japanese radiology reports in our hospital were used, of which 788 (0.87%) were actionable. We evaluated four methods, statistical machine learning with logistic regression (LR) and with gradient boosting decision tree (GBDT), and deep learning with a bidirectional long short-term memory (LSTM) model and a publicly available Japanese BERT model. Each method was used with two different inputs, radiology reports alone and pairs of order information and radiology reports. Thus, eight experiments were conducted to examine the performance. RESULTS: Without order information, BERT achieved the highest area under the precision-recall curve (AUPRC) of 0.5138, which showed a statistically significant improvement over LR, GBDT, and LSTM, and the highest area under the receiver operating characteristic curve (AUROC) of 0.9516. Simply coupling the order information with the radiology reports slightly increased the AUPRC of BERT but did not lead to a statistically significant improvement. This may be due to the complexity of clinical decisions made by radiologists. CONCLUSIONS: BERT was assumed to be useful to detect actionable reports. More sophisticated methods are required to use order information effectively.

Unsupervised Deep Anomaly Detection in Chest Radiographs.

The purposes of this study are to propose an unsupervised anomaly detection method based on a deep neural network (DNN) model, which requires only normal images for training, and to evaluate its performance with a large chest radiograph dataset. We used the auto-encoding generative adversarial network (α-GAN) framework, which is a combination of a GAN and a variational autoencoder, as a DNN model. A total of 29,684 frontal chest radiographs from the Radiological Society of North America Pneumonia Detection Challenge dataset were used for this study (16,880 male and 12,804 female patients; average age, 47.0 years). All these images were labeled as "Normal," "No Opacity/Not Normal," or "Opacity" by board-certified radiologists. About 70% (6,853/9,790) of the Normal images were randomly sampled as the training dataset, and the rest were randomly split into the validation and test datasets in a ratio of 1:2 (7,610 and 15,221). Our anomaly detection system could correctly visualize various lesions including a lung mass, cardiomegaly, pleural effusion, bilateral hilar lymphadenopathy, and even dextrocardia. Our system detected the abnormal images with an area under the receiver operating characteristic curve (AUROC) of 0.752. The AUROCs for the abnormal labels Opacity and No Opacity/Not Normal were 0.838 and 0.704, respectively. Our DNN-based unsupervised anomaly detection method could successfully detect various diseases or anomalies in chest radiographs by training with only the normal images.

Therapeutic effect of portal vein stenting for portal vein stenosis after upper-abdominal surgery.

BACKGROUND: The therapeutic effect of portal vein (PV) stenting for PV stenosis following nontransplant hepato-pancreato-biliary (HPB) surgery has not been fully investigated. METHODS: Changes in portal venous pressure (PVP) gradient before and after stenting, complications, symptomatic improvement, and stent patency were evaluated. RESULTS: We identified 14 consecutive patients undergoing PV stenting for malignant (n = 8) and benign (n = 6) PV stenosis. Signs of PV stenosis were composed of refractory ascites in 6 patients, varices with hemorrhagic tendencies in 5, and abnormal liver function in 5. The median PVP gradient after PV stenting was 3.0 cm H2O (range, 1.5-3.0), which was significantly smaller than that before PV stenting (median, 15 cm H2O [range, 2.5-25]; P < 0.01). Thirteen out of 14 (93%) achieved clinical success with symptomatic improvement, except one patient with sustained refractory ascites because of peritoneal seeding. During the median follow-up time of 7.3 months (range, 1.0-87), stent occlusion occurred in two patients (14%) because of intrastent tumor growth. The 1-year cumulative stent patency rate was 76% in the entire cohort. CONCLUSIONS: Based on durable effect on patency, we deemed PV stenting for PV stenosis after HPB surgery to be safe and beneficial for improving symptoms.

Clinical practice guidelines for hepatocellular carcinoma: The Japan Society of Hepatology 2017 (4th JSH-HCC guidelines) 2019 update.

The fourth version of Clinical Practice Guidelines for Hepatocellular Carcinoma was revised by the Japan Society of Hepatology, according to the methodology of evidence-based medicine and partly to the Grading of Recommendations Assessment, Development, and Evaluation system, which was published in October 2017 in Japanese. New or revised recommendations were described, herein, with a special reference to the surveillance, diagnostic, and treatment algorithms.

Sarcopenia, intramuscular fat deposition, and visceral adiposity independently predict the outcomes of hepatocellular carcinoma.

BACKGROUND & AIMS: Obesity defined by body mass index (BMI) significantly increases the risk of hepatocellular carcinoma (HCC). In contrast, not only obesity but also underweight is associated with poor prognosis in patients with HCC. Differences in body composition rather than BMI were suggested to be true determinants of prognosis. However, this hypothesis has not been demonstrated conclusively. METHODS: We measured skeletal muscle index (SMI), mean muscle attenuation (MA), visceral adipose tissue index, subcutaneous adipose tissue index, and visceral to subcutaneous adipose tissue area ratios (VSR) via computed tomography in a large-scale retrospective cohort of 1257 patients with different stages of HCC, and comprehensively analyzed the impact of body composition on the prognoses. RESULTS: Among five body composition components, low SMI (called sarcopenia), low MA (called intramuscular fat [IMF] deposition), and high VSR (called visceral adiposity) were significantly associated with mortality, independently of cancer stage or Child-Pugh class. A multivariate analysis revealed that sarcopenia (hazard ratio [HR], 1.52; 95% confidence interval [CI], 1.18-1.96; p=0.001), IMF deposition (HR, 1.34; 95% CI, 1.05-1.71; p=0.020), and visceral adiposity (HR, 1.35; 95% CI, 1.09-1.66; p=0.005) but not BMI were significant predictors of survival. The prevalence of poor prognostic body composition components was significantly higher in underweight and obese patients than in normal weight patients. CONCLUSIONS: Sarcopenia, IMF deposition, and visceral adiposity independently predict mortality in patients with HCC. Body composition rather than BMI is a major determinant of prognosis in patients with HCC.

Performance changes due to differences among annotating radiologists for training data in computerized lesion detection.

PURPOSE: The quality and bias of annotations by annotators (e.g., radiologists) affect the performance changes in computer-aided detection (CAD) software using machine learning. We hypothesized that the difference in the years of experience in image interpretation among radiologists contributes to annotation variability. In this study, we focused on how the performance of CAD software changes with retraining by incorporating cases annotated by radiologists with varying experience. METHODS: We used two types of CAD software for lung nodule detection in chest computed tomography images and cerebral aneurysm detection in magnetic resonance angiography images. Twelve radiologists with different years of experience independently annotated the lesions, and the performance changes were investigated by repeating the retraining of the CAD software twice, with the addition of cases annotated by each radiologist. Additionally, we investigated the effects of retraining using integrated annotations from multiple radiologists. RESULTS: The performance of the CAD software after retraining differed among annotating radiologists. In some cases, the performance was degraded compared to that of the initial software. Retraining using integrated annotations showed different performance trends depending on the target CAD software, notably in cerebral aneurysm detection, where the performance decreased compared to using annotations from a single radiologist. CONCLUSIONS: Although the performance of the CAD software after retraining varied among the annotating radiologists, no direct correlation with their experience was found. The performance trends differed according to the type of CAD software used when integrated annotations from multiple radiologists were used.

Interobserver variability in the CT assessment of honeycombing in the lungs.

PURPOSE: To quantify observer agreement and analyze causes of disagreement in identifying honeycombing at chest computed tomography (CT). MATERIALS AND METHODS: The institutional review board approved this multiinstitutional HIPAA-compliant retrospective study, and informed patient consent was not required. Five core study members scored 80 CT images with a five-point scale (5 = definitely yes to 1 = definitely no) to establish a reference standard for the identification of honeycombing. Forty-three observers from various subspecialties and geographic regions scored the CT images by using the same scoring system. Weighted κ values of honeycombing scores compared with the reference standard were analyzed to investigate intergroup differences. Images were divided into four groups to allow analysis of imaging features of cases in which there was disagreement: agreement on the presence of honeycombing, agreement on the absence of honeycombing, disagreement on the presence of honeycombing, and other (none of the preceding three groups applied). RESULTS: Agreement of scores of honeycombing presence by 43 observers with the reference standard was moderate (Cohen weighted κ values: 0.40-0.58). There were no significant differences in κ values among groups defined by either subspecialty or geographic region (Tukey-Kramer test, P = .38 to >.99). In 29% of cases, there was disagreement on identification of honeycombing. These cases included honeycombing mixed with traction bronchiectasis, large cysts, and superimposed pulmonary emphysema. CONCLUSION: Identification of honeycombing at CT is subjective, and disagreement is largely caused by conditions that mimic honeycombing.

Effects of deep learning on radiologists' and radiology residents' performance in identifying esophageal cancer on CT.

OBJECTIVE: To investigate the effectiveness of a deep learning model in helping radiologists or radiology residents detect esophageal cancer on contrast-enhanced CT images. METHODS: This retrospective study included 250 and 25 patients with and without esophageal cancer, respectively, who underwent contrast-enhanced CT between December 2014 and May 2021 (mean age, 67.9 ± 10.3 years; 233 men). A deep learning model was developed using data from 200 and 25 patients with esophageal cancer as training and validation data sets, respectively. The model was then applied to the test data set, consisting of additional 25 and 25 patients with and without esophageal cancer, respectively. Four readers (one radiologist and three radiology residents) independently registered the likelihood of malignant lesions using a 3-point scale in the test data set. After the scorings were completed, the readers were allowed to reference to the deep learning model results and modify their scores, when necessary. RESULTS: The area under the curve (AUC) of the deep learning model was 0.95 and 0.98 in the image- and patient-based analyses, respectively. By referencing to the deep learning model results, the AUCs for the readers were improved from 0.96/0.93/0.96/0.93 to 0.97/0.95/0.99/0.96 (p = 0.100/0.006/<0.001/<0.001, DeLong's test) in the image-based analysis, with statistically significant differences noted for the three less-experienced readers. Furthermore, the AUCs for the readers tended to improve from 0.98/0.96/0.98/0.94 to 1.00/1.00/1.00/1.00 (p = 0.317/0.149/0.317/0.073, DeLong's test) in the patient-based analysis. CONCLUSION: The deep learning model mainly helped less-experienced readers improve their performance in detecting esophageal cancer on contrast-enhanced CT. ADVANCES IN KNOWLEDGE: A deep learning model could mainly help less-experienced readers to detect esophageal cancer by improving their diagnostic confidence and diagnostic performance.

Multimodal imaging findings of intrahepatic cholangiocarcinoma arising from a biliary adenofibroma: a case report with radiological-pathological correlation.

PURPOSE: Biliary adenofibroma is a solid microcystic epithelial neoplasm in the liver, comprising microcystic and tubuloacinar glandular tissues lined by a non-mucin secreting biliary epithelium and supported by a fibrous stroma. It is an extremely rare benign tumor with potential for malignant transformation. Herein, we report the case of a 64-year-old woman diagnosed with intrahepatic cholangiocarcinoma arising from biliary adenofibroma. METHODS: Imaging studies revealed a tumor of 50 mm diameter, consisting of two components in S1 of the liver. The ventral portion of the tumor showed an ill-defined mass with early peripheral and gradual centripetal enhancement invading to the middle hepatic vein on computed tomography (CT), diffusion restriction on magnetic resonance images, and high fluorine-18-2-deoxy-D-glucose (FDG) uptake on positron emission tomography, like conventional intrahepatic cholangiocarcinoma. The dorsal portion showed a well-defined and low-attenuated mass with heterogeneous early enhancement and partial wash-out on CT, marked hyperintensity on heavily T2-weighted images, and low FDG uptake. The patient subsequently underwent extended left hepatectomy. RESULTS: Pathologically, the former was diagnosed as cholangiocarcinoma and the latter as biliary adenofibroma. We discuss the radiological-pathological correlation of the tumor with a literature review. CONCLUSION: Preoperative diagnosis of biliary adenofibroma is extremely challenging; however, clinically, it is crucial not to miss the presence of malignant findings.

Intraductal oncocytic papillary neoplasm of the pancreas: clinical and radiological features compared to those of intraductal papillary mucinous neoplasm.

PURPOSE: This study aimed to characterize the clinical and imaging findings of intraductal oncocytic papillary neoplasm of the pancreas (IOPN-P) compared to those of intraductal papillary mucinous adenoma/carcinoma (IPMA/IPMC). METHODS: This multi-institutional retrospective study reviewed the clinical, imaging, and pathological findings of 21 patients with pathologically proven IOPN-P. Twenty-one computed tomography (CT) and magnetic resonance imaging, and seven 18F-fluorodeoxyglucose (FDG)-positron emission tomography were performed before surgery. The following findings were evaluated: preoperative blood test results, lesion size and location, pancreatic duct diameter, contrast-enhancement effect, bile duct and peripancreatic invasion, maximum standardized uptake (SUVmax) value, and pathological stromal invasion. RESULTS: Serum carcinoembryonic antigen (CEA) and cancer antigen 19-9 (CA19-9) levels were significantly higher in the IPMN/IPMC group than in the IOPN-P group. Except in one patient, IOPN-P showed multifocal cystic lesions with solid components or a tumor in the main pancreatic duct (MPD) with dilatation. IOPN-P had a higher frequency of solid parts and a lower frequency of downstream MPD dilatation than IPMA. IPMC showed smaller overall cyst size, more radiological peripancreatic invasion, and worse recurrence-free and overall survival than IOPN-P. The average SUVmax value of IOPN-P was 7.5. Pathologically, 17 of the 21 IOPN-Ps had a malignant component, and six showed stromal invasion. CONCLUSION: IOPN-P shows cystic-solid lesions similar to IPMC but has lower serum CEA and CA19-9 levels, larger overall cyst size, lower frequency of peripancreatic invasion, and more favorable prognosis than IPMC. Moreover, the high FDG uptake by IOPN-Ps may be a characteristic finding of this study.

Vertebral artery variations at the C1-2 level diagnosed by magnetic resonance angiography.

INTRODUCTION: The craniovertebral junction is clinically important. The vertebral artery (VA) in its several variations runs within this area. We report the prevalence of these VA variations on magnetic resonance angiography (MRA). METHODS: We retrospectively reviewed MRA images, obtained using two 1.5-T imagers, of 2,739 patients, and paid special attention to the course and branching of the VA at the level of the C1-2 vertebral bodies. RESULTS: There were three types of VA variation at the C1-2 level: (1) persistent first intersegmental artery (FIA), (2) VA fenestration, and (3) posterior inferior cerebellar artery (PICA) originating from the C1/2 level. The overall prevalence of these three variations was 5.0%. There was no laterality in frequency, but we found female predominance (P < 0.05). We most frequently observed the persistent FIA (3.2%), which was sometimes bilateral. We found VA fenestration (0.9%) and PICA of C1/2 origin (1.1%) with almost equal frequency. Two PICAs of C1/2 origin had no normal VA branch. CONCLUSIONS: We frequently observed VA variations at the C1-2 level and with female predominance. The persistent FIA was most prevalent and sometimes seen bilaterally. Preoperative identification of these variations in VA is necessary to avoid complications during surgery at the craniovertebral junction.

Prognosis of hepatocellular carcinoma with portal vein tumor thrombus:assessment based on clinical and computer tomography characteristics.

Patients with hepatocellular carcinoma (HCC) complicated by portal vein tumor thrombus (PVTT) have an extremely poor prognosis. It is important to select adequate therapeutic options based on reliable prognostic factors using imaging studies and clinical data. Prognostic factors were analyzed in patients with HCC with PVTT in the first branch or main trunk of the portal vein. From 2000 to 2007, 107 consecutive patients with HCC with PVTT in the major portal vein were reviewed, and diagnostic images and clinical characteristics were retrospectively observed. Thirty-eight possible prognostic factors for survival were analyzed by the log-rank test and multivariate analysis using Cox's proportional hazards model. Median overall survival was 14 months following PVTT diagnosis. Survival rates at 6 months, 1, 2, and 3 years were 72.1%, 52.6%, 32.6%, and 29.6%, respectively. Independent prognostic factors for longer survival included:patient age < 65 years, Child-Pugh classification A/B, PVTT treatment, accumulation of Lipiodol in the PVTT after TACE, initial radical treatment for HCC, HCC located in a single lobe of the liver, and no invasion of HCC to the hepatic vein or bile duct. Survival was associated with liver function, tumor extension, and treatment for HCC and PVTT.

Successful transcatheter arterial embolization for a massive hemothorax caused by acupuncture.

Acupuncture is an alternative treatment for a variety of diseases, and serious complications are rare. We report a case of transcatheter arterial embolization performed in a patient with a massive hemothorax after acupuncture treatment. A 36-year-old woman with no previous medical history was admitted to our hospital with left back pain and respiratory distress after acupuncture treatment. Contrast-enhanced computed tomography showed a left hemothorax and leakage of contrast medium, which was considered to result from an injury to the second intercostal artery, caused by acupuncture treatment. Transcatheter arterial embolization successfully stopped the bleeding, and the hematoma was thoracoscopically removed. No rebleeding was observed 6 months after treatment.

A Case of Preeclampsia with Uterine Necrosis after Uterine Artery Embolization for Postpartum Hemorrhage.

Uterine necrosis is a rare complication in uterine artery embolization (UAE) for postpartum hemorrhage (PPH). Preeclampsia (PE) is a condition characterized with systemic endothelial damage and intravascular volume depletion. Whether a patient with PE is at high risk for uterine necrosis after UAE for PPH has been unknown. A 30-year-old primipara woman was diagnosed with PE based on hypertension and proteinuria during delivery. UAE was performed for PPH after forceps delivery. After UAE, the patient presented with pleural effusion and massive ascites as well as persistent fever unresponsive to antibiotics. Ultrasonography and contrast-enhanced magnetic resonance imaging (MRI) led to the diagnosis of uterine necrosis, for which we performed total laparoscopic hysterectomy. It should be kept in mind that patients with PE associated with massive ascites may be at high risk for uterine necrosis after UAE due to decreased uterine perfusion. Therefore, it is important to pay attention to persistent symptoms such as fever and abdominal pain after UAE to diagnose uterine necrosis.

Schwannoma in and around the porta hepatis: radiological features in eight cases and literature review.

PURPOSE: Schwannomas in and around the porta hepatis (porta hepatic schwannomas) are rare benign tumors easily misdiagnosed as other pathologies, including malignancies. We aimed to evaluate their imaging features on ultrasonography, computed tomography (CT), magnetic resonance imaging (MRI), and 18F-fluorodeoxyglucose-positron emission tomography/CT (FDG-PET/CT). METHODS: We performed a multi-institutional retrospective study by reviewing the clinical and imaging findings of pathologically proven eight porta hepatic schwannomas (mean age, 55 years; range, 38-80 years; one male and seven females). Preoperative imaging included three ultrasonography, eight CT, eight MRI, and two FDG-PET/CT. RESULTS: All patients were asymptomatic. The mean tumor size was 61.9 mm (range, 30-180 mm), and all tumors demonstrated well-defined lesions on ultrasonography and their solid components showed soft tissue attenuation on non-contrast CT. MRI showed two distinct components in all cases: the component with T1-weighted hypointensities and T2-weighted hyperintensities with poor enhancement (suggestive of Antoni B histology); the component with T2-weighted hypointensities with gradually increasing enhancement (suggestive of Antoni A histology), resulting in a heterogeneous pattern on post-contrast CT or MRI (8/8, 100%). The separated deviation of surrounding bile ducts and vessels without obstruction allowed the recognition of extrahepatic localization and their benign nature. A ginger root-like morphology (2/8, 25%) seemed to be suggestive of extension along the Glisson's sheath, although this finding was not seen frequently. CONCLUSION: Recognizing imaging features such as extrahepatic location, benign nature with internal structures suggestive of Antoni A/B histology, and characteristic tumor extension may provide key diagnostic clues for porta hepatic schwannomas.

Deep Learning Algorithm for Fully Automated Detection of Small (≤4 cm) Renal Cell Carcinoma in Contrast-Enhanced Computed Tomography Using a Multicenter Database.

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.