Liver imaging features by convolutional neural network to predict the metachronous liver metastasis in stage I-III colorectal cancer patients based on preoperative abdominal CT scan.

BACKGROUND: Introducing deep learning approach to medical images has rendered a large amount of un-decoded information into usage in clinical research. But mostly, it has been focusing on the performance of the prediction modeling for disease-related entity, but not on the clinical implication of the feature itself. Here we analyzed liver imaging features of abdominal CT images collected from 2019 patients with stage I - III colorectal cancer (CRC) using convolutional neural network (CNN) to elucidate its clinical implication in oncological perspectives. RESULTS: CNN generated imaging features from the liver parenchyma. Dimension reduction was done for the features by principal component analysis. We designed multiple prediction models for 5-year metachronous liver metastasis (5YLM) using combinations of clinical variables (age, sex, T stage, N stage) and top principal components (PCs), with logistic regression classification. The model using "1st PC (PC1) + clinical information" had the highest performance (mean AUC = 0.747) to predict 5YLM, compared to the model with clinical features alone (mean AUC = 0.709). The PC1 was independently associated with 5YLM in multivariate analysis (beta = - 3.831, P < 0.001). For the 5-year mortality rate, PC1 did not contribute to an improvement to the model with clinical features alone. For the PC1, Kaplan-Meier plots showed a significant difference between PC1 low vs. high group. The 5YLM-free survival of low PC1 was 89.6% and the high PC1 was 95.9%. In addition, PC1 had a significant correlation with sex, body mass index, alcohol consumption, and fatty liver status. CONCLUSION: The imaging features combined with clinical information improved the performance compared to the standardized prediction model using only clinical information. The liver imaging features generated by CNN may have the potential to predict liver metastasis. These results suggest that even though there were no liver metastasis during the primary colectomy, the features of liver imaging can impose characteristics that could be predictive for metachronous liver metastasis.

The exploration of feature extraction and machine learning for predicting bone density from simple spine X-ray images in a Korean population.

OBJECTIVE: Osteoporosis is hard to detect before it manifests symptoms and complications. In this study, we evaluated machine learning models for identifying individuals with abnormal bone mineral density (BMD) through an analysis of spine X-ray features extracted by deep learning to alert high-risk osteoporosis populations. MATERIALS AND METHODS: We retrospectively used data obtained from health check-ups including spine X-ray and dual-energy X-ray absorptiometry (DXA). Consecutively, we selected people with normal and abnormal bone mineral density. From the regions of interest of X-ray images, deep convolutional networks were used to generate image features. We designed prediction models for abnormal BMD using the image features trained by machine learning classification algorithms. The performances of each model were evaluated. RESULTS: From 334 participants, 170 images of abnormal (T scores < - 1.0 standard deviations (SD)) and 164 of normal BMD (T scores > = - 1.0 SD) were used for analysis. We found that a combination of feature extraction by VGGnet and classification by random forest based on the maximum balanced classification rate (BCR) yielded the best performance in terms of the area under the curve (AUC) (0.74), accuracy (0.71), sensitivity (0.81), specificity (0.60), BCR (0.70), and F1-score (0.73). CONCLUSION: In this study, we explored various machine learning algorithms for the prediction of BMD using simple spine X-ray image features extracted by three deep learning algorithms. We identified the combination for the best performance in predicting high-risk populations with abnormal BMD.

Dual-enhanced cardiac CT for detection of left atrial appendage thrombus in patients with stroke: a prospective comparison study with transesophageal echocardiography.

BACKGROUND AND PURPOSE: A noninvasive method with high reliability and accuracy comparable to transesophageal echocardiography for identification of left atrial appendage thrombus would be of significant clinical value. The aim of this study was to assess the diagnostic performance of a dual-enhanced cardiac CT protocol for detection of left atrial appendage thrombi and for differentiation between thrombus and circulatory stasis in patients with stroke. METHODS: We studied 83 consecutive patients with stroke (56 men and 27 women; mean age, 62.6 years) who had high risk factors for thrombus formation and had undergone both dual-source CT and transesophageal echocardiography within a 3-day period. CT was performed with prospective electrocardiographic gating, and scanning began 180 seconds after the test bolus. RESULTS: Among the 83 patients, a total of 13 thrombi combined with spontaneous echo contrast and 14 spontaneous echo contrasts were detected by transesophageal echocardiography. All 13 thrombi combined with spontaneous echo contrast were correctly diagnosed on CT. Using transesophageal echocardiography as the reference standard, the overall sensitivity and specificity of CT for the detection of thrombi and circulatory stasis in the left atrial appendage were 96% (95% CI, 78% to 99%), and 100% (95% CI, 92% to 100%), respectively. On CT, the mean left atrial appendage/ascending aorta Hounsfield unit ratios were significantly different between thrombus and circulatory stasis (0.15 Hounsfield unit versus 0.27 Hounsfield unit, P=0.001). The mean effective radiation dose was 3.11 mSv. CONCLUSIONS: Dual-enhanced cardiac CT with prospective electrocardiographic gating is a noninvasive and sensitive modality for detecting left atrial appendage thrombus with an acceptable radiation dose.

CT fluoroscopy-guided lung biopsy versus conventional CT-guided lung biopsy: a prospective controlled study to assess radiation doses and diagnostic performance.

OBJECTIVE: We evaluated radiation doses, complication rates, and diagnostic accuracy for CT-guided percutaneous needle aspiration biopsy (NAB) procedures of pulmonary lesions performed with or without fluoroscopic guidance. METHODS: A total of 142 patients were prospectively enrolled to receive CT-guided NAB with (Group I, n = 72) or without (Group II, n = 70) fluoroscopic guidance. Outcome measurements were patient and doctor radiation dose, and complication rate. Sensitivity, specificity and accuracy were calculated based on 123 NAB results. RESULTS: The mean estimated effective patient radiation dose was 6.53 mSv in Group I and 2.72 mSv in Group II (p < 0.001). The mean estimated effective doctor dose was 0.054 mSv in Group I and 0.029 mSv in Group II (p < 0.001). The complication rate was significantly different between the two groups (13.4% versus 31.4%, p = 0.012). Sensitivity, specificity and accuracy for diagnosing pulmonary lesions were 97.8%, 100% and 98.4% in group I and 95.3%, 100% and 89.5% in group II (p > 0.05). CONCLUSIONS: CT fluoroscopy-guided NAB of pulmonary lesions provides high diagnostic accuracy and can be performed with significantly fewer complications. However, radiation exposure to both patient and doctor were significantly higher than conventional CT-guided NAB.

Computed tomographic fluoroscopy-guided needle aspiration biopsy as a second biopsy technique after indeterminate transbronchial biopsy results for pulmonary lesions: comparison with second transbronchial biopsy.

OBJECTIVE: The purpose of this study was to compare the diagnostic performance of computed tomographic (CT) fluoroscopy-guided percutaneous transthoracic needle aspiration biopsy (NAB) and transbronchial lung biopsy (TBLB) after indeterminate bronchoscopy in patients with suspected malignant pulmonary lesions. METHODS: We included 77 patients who underwent CTF-NAB (n = 53) or TBLB (n = 24) as a second biopsy for pulmonary lesions because of inconclusive pathologic results on initial TBLB. Sensitivity, specificity, and diagnostic accuracy were calculated and compared between the 2 groups using the Fisher exact test. Sensitivity and diagnostic accuracy were also compared according to lesion depth (central vs peripheral), lesion location (upper vs lower), and lesion size (<2 vs 2-3 vs >3 cm). RESULTS: There were 50 (65%) malignant and 27 (35%) benign lesions. The overall sensitivity, specificity, and accuracy for diagnosing pulmonary lesions were 84%, 100%, and 91% for NAB and 50%, 100%, and 63% for TBLB. The sensitivity and accuracy for diagnosing pulmonary lesions were significantly different between the 2 groups (P = 0.019, and P = 0.008). The sensitivity and accuracy of TBLB for diagnosing lesions was significantly different according to the lesion size (P = 0.025, and P = 0.048). CONCLUSION: A second biopsy using CT fluoroscopy-guided NAB is a useful diagnostic modality for exact diagnosis of pulmonary lesions in cases of inconclusive pathologic results on initial TBLB.

The risk of atrial fibrillation in patients with non-alcoholic fatty liver disease and a high hepatic fibrosis index.

Previous epidemiological studies focusing on the association between liver disease and atrial fibrillation (AF) show interesting but inconsistent findings. Patients with liver disease have a higher AF risk; however, it is unknown whether the liver fibrosis index can predict AF risk. The medical records of a healthy population undergoing routine health examinations at Healthcare System Gangnam Center, Seoul National University Hospital, were reviewed retrospectively. After excluding subjects with a history of liver disease and known cardiovascular disease, 74,946 subjects with nonalcoholic fatty liver disease (NAFLD) were evaluated. The mean age was 51 ± 11 years, and 71.9% were male. AF was found in 380 (0.5%) subjects. Using univariate analyses, age, male sex, body mass index, hypertension, and diabetes were significantly associated with AF. The fibrosis 4 index (FIB 4) showed significant correlations with AF [unadjusted odds ratio (OR) 3.062 and 95% confidence interval (CI) 2.605-3.600, p = 0.000; adjusted OR 2.255 and 95% CI 1.744-2.915, p = 0.000, with cardiometabolic risk factors adjusted]. In conclusion, NAFLD subjects with higher FIB 4 were associated with increased AF risk. The noninvasive determination of liver fibrosis indices can have clinical implications on the early identification of NAFLD in patients at risk for AF.

Metastasis versus focal eosinophilic infiltration of the liver in patients with extrahepatic abdominal cancer: an evaluation with gadobenate dimeglumine-enhanced magnetic resonance imaging.

PURPOSE: To determine the performance of gadobenate dimeglumine (Gd-BOPTA)-enhanced magnetic resonance imaging (MRI) to differentiate focal eosinophilic infiltration (FEI) from metastasis. MATERIALS AND METHODS: Gadobenate dimeglumine-enhanced MRI conducted in 38 patients (lesions of 43 metastases and 33 FEIs) with extrahepatic abdominal carcinoma were reviewed retrospectively. Images were divided into 2 sets. Set 1 was composed of precontrast MRI with dynamic phase images. Set 2 included hepatobiliary phase images in addition to the images of set 1. Two blinded radiologists independently categorized each set of images into 5 grades to differentiate metastases from FEI. The area under the receiver operating characteristic curve (Az) was calculated for each set of images. The observers evaluated the shape, signal intensity (SI), and enhancement pattern of the lesions in consensus. RESULTS: The Az values for set 2 images (reader 1: 0.976, and reader 2: 0.743) were greater than those of set 1 (0.961 and 0.709), without statistical significance (P = 0.470 and 0.223). Target appearance was seen in 27 of 43 (63%) metastases on hepatobiliary phase imaging, but none on the FEI. Twenty-five (76%) of the 33 FEIs showed intermingled hypo-SI and iso-SI on hepatobiliary phase images with changing size and shape compared to the T2-weighted images, whereas none of the metastases did. CONCLUSION: Gadobenate dimeglumine-enhanced MRI may assist in the differentiation of FEI from metastasis by highlighting the distinctive features between them, even without hepatobiliary phase imaging.

US-guided vacuum-assisted biopsy of microcalcifications in breast lesions and long-term follow-up results.

OBJECTIVE: To evaluate the diagnostic accuracy of the use of an ultrasonography (US)-guided vacuum-assisted biopsy for microcalcifications of breast lesions and to evaluate the efficacy of the use of US-guided vacuum-assisted biopsy with long-term follow-up results. MATERIALS AND METHODS: US-guided vacuum-assisted biopsy cases of breast lesions that were performed between 2002 and 2006 for microcalcifications were retrospectively reviewed. A total of 62 breast lesions were identified where further pathological confirmation was obtained or where at least two years of mammography follow-up was obtained. These lesions were divided into the benign and malignant lesions (benign and malignant group) and were divided into underestimated group and not-underestimated lesions (underestimated and not-underestimated group) according to the diagnosis after a vacuum-assisted biopsy. The total number of specimens that contained microcalcifications was analyzed and the total number of microcalcification flecks as depicted on specimen mammography was analyzed to determine if there was any statistical difference between the groups. RESULTS: There were no false negative cases after more than two years of follow-up. Twenty-nine lesions were diagnosed as malignant (two invasive carcinomas and 27 carcinoma in situ lesions). Two of the 27 carcinoma in situ lesions were upgraded to invasive cancers after surgery. Among three patients diagnosed with atypical ductal hyperplasia, the diagnosis was upgraded to a ductal carcinoma in situ after surgery in one patient. There was no statistically significant difference in the number of specimens with microcalcifications and the total number of microcalcification flecks between the benign group and malignant group of patients and between the underestimated group and not-underestimated group of patients. CONCLUSION: US-guided vacuum-assisted biopsy can be an effective alternative to stereotactic-guided vacuum-assisted biopsy in cases where microcalcifications are visible with the use of high-resolution US.

Epicardial fat reflects arterial stiffness: assessment using 256-slice multidetector coronary computed tomography and cardio-ankle vascular index.

AIM: The cardio-ankle vascular index (CAVI) reflects overall arterial stiffness from the aorta to the ankle, independent of blood pressure. We aimed to investigate the association of fat burden assessed by visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT) and epicardial adipose tissue (EAT) with CAVI in an asymptomatic population. METHODS: A total of 260 asymptomatic Korean individuals who had CAVI, abdominal computed tomography (CT) and coronary CT were evaluated retrospectively. The VAT, SAT, EAT and SAT to VAT ratio (SVR) were measured and assessed for correlation with CAVI. RESULTS: Different fat compartments showed different correlations with arterial stiffness as assessed by CAVI. The amount of fat measured by VAT (r= 0.129, p= 0.037), EAT (r=0.193, p= 0.002) and SVR (r=-0.168, p= 0.007) showed a significant correlation with CAVI, whereas the amount of total abdominal fat and SAT did not (p= 0.261 and p= 0.434 respectively). From step-wise multivariate regression analysis including age, pulse pressure, fasting blood sugar level, VAT, SVR and EAT, EAT (p= 0.036) and age (p<0.001) showed significant associations with CAVI. When quartiles of CAVI were assessed, EAT showed serial increment, whereas SVR showed a stepwise decrease from the first quartile to fourth quartile of CAVI (p=0.041). CONCLUSION: VAT, EAT and SVR, which reflect metabolic risk, have shown significant correlations with arterial stiffness measured by CAVI. EAT showed an independent association with arterial stiffness after adjusting for covariables by multivariable correlation analysis. Among the different parameters reflecting fat burden, EAT showed the strongest correlation with CAVI.

Low-dose electrocardiography synchronized nonenhanced computed tomography for assessing left atrium and pulmonary veins before radiofrequency catheter ablation for atrial fibrillation.

Recently contrast-enhanced cardiac computed tomography (CT) was found to be useful for imaging the left atrium and pulmonary veins (PVs) before radiofrequency catheter ablation in patients with atrial fibrillation. However, the risks of contrast agent in patients with impaired renal function must be considered. We investigated the accuracy of low-dose electrocardiographically synchronized nonenhanced cardiac CT (NECT) for identifying PV anatomy. One hundred eight consecutive patients who underwent cardiac CT before radiofrequency catheter ablation of atrial fibrillation were included. Nonenhanced cardiac computed tomogram was retrospectively evaluated for each patient by 2 radiologists for the following PV anatomy: conventional pattern, conjoined ostium, and accessory PVs with number and location. Sensitivity and specificity for variations in PVs were calculated using contrast-enhanced cardiac computed tomogram as the reference standard. Detection rates for each variation were also calculated. Twenty-one right PV (RPV) variations and 11 left PV (LPV) variations were observed. NECT showed a high diagnostic performance in detecting variations in PVs for the 2 observers. For RPV variations overall sensitivity was 97.6% and specificity was 96.6%. For LPV variations overall sensitivity was 90.9% and specificity was 97.9%. Overall detection rates for variation between the 2 observers were 97.1% for accessory RPV from the right middle lobe, 100% for 4 ostia with accessory RPV from the right middle lobe and accessory RPV from the superior segment of the right lower lobe, 100% for accessory RPV from the superior segment of the right lower lobe, 88.9% for conjoined ostium of the LPV, and 100% for accessory LPV from the left lingular segment. In conclusion, variations in PV anatomy were detected with great accuracy by NECT.

Analysis of tumor markers in the cytological fluid obtained from computed tomography-guided needle aspiration biopsy for the diagnosis of non-small cell lung cancer.

PURPOSE: The aim of this study was to prospectively assess whether analysis of the tumor markers cytokeratin 19 fragments (CYFRA 21-1), carcinoembryonic antigen (CEA), and squamous cell carcinoma (SCC) antigen in cytological fluid can improve the performance of computed tomography (CT)-guided needle aspiration biopsy (NAB) in the diagnosis of non-small cell lung cancer (NSCLC). METHODS: A total of 100 patients (men:women = 41:59, mean age: 63 years) with suspected malignant pulmonary lesions were prospectively enrolled for CT-guided NAB procedures. Levels of CYFRA 21-1, CEA, and SCC in the cytological fluid were measured by immunoradiometric assays. The cutoff value for tumor markers was selected on the basis of best accuracy through receiver operating characteristic curves. The sensitivity and areas under the curve (AUC) of NAB alone were compared with those of NAB combined with cytological tumor markers (CYFRA 21-1, CEA, and SCC). RESULTS: Among 100 patients, 71 (71%) had NSCLC and 29 (29%) had benign lesions. The sensitivity, specificity, and accuracy for diagnosing NSCLC were 85.7%, 100%, and 89%, respectively, for NAB alone. The sensitivity increased significantly for NAB combined with a tumor marker compared with NAB alone (100% for CYFRA 21-1, 92.9% for CEA, and 94.2% for SCC; p = 0.001, p = 0.025, and p = 0.014, respectively). The AUC of NAB with CYFRA 21-1 was significantly larger than the AUC of NAB alone (p = 0.001). CONCLUSION: Evaluation of tumor markers CYFRA 21-1, CEA, and SCC in the cytological fluid can improve the diagnostic performance of CT-guided NAB for NSCLC. Of these markers, CYFRA 21-1 is the most useful cytological tumor marker.

Radiologic findings of lung lobe torsion in reconstructed multidetector computed tomography image lead to early detection.

Torsion of the remaining lung lobe after lobectomy is a rare complication, and sometimes, diagnosis can be difficult. But early detection is important to prevent resection and fatal complication. We present a case of left lower lobe torsion with reconstructed image of multidetector computed tomography (MDCT) findings, which propose a possibility of early detection on retrospective review after surgical reduction. Multiplanar reconstruction of MDCT image, 3-dimensional CT angiography, and CT bronchography could provide important diagnostic clues.

Pseudoaneurysm of the inferior thyroid artery presenting as a thyroid nodule.

A 63-year-old woman complained of a growing neck mass. Sonography revealed a solid and cystic mass in the right thyroid lobe. Color Doppler study and computed tomography revealed anechoic cystic portions within the thyroid nodule to be a pseudoaneurysm arising from the right inferior thyroid artery. The only identifiable risk factor was a past thyroid needle biopsy performed approximately 5 years previously according to the patient's recollection. Right lobectomy was performed and the pseudoaneurysm was treated without complication.

Single breath-hold multiarterial dynamic MRI of the liver at 3T using a 3D fat-suppressed keyhole technique.

PURPOSE: To evaluate the feasibility of single breath-hold, multiarterial MRI of the liver using the THRIVE-CENTRA-keyhole technique. MATERIALS AND METHODS: Twenty-eight patients with 63 focal hepatic lesions underwent liver MR examinations that included the three-dimensional THRIVE-CENTRA-keyhole sequence. Three or six phases were obtained for arterial phase scanning during a single breath-hold. Central k-space data were collected for each phase but the remaining peripheral k-space data were collected only once. The enhancement pattern of each hepatic lesion was analyzed according to the specific diagnosis. RESULTS: Hepatocellular carcinomas (n = 24) enhancement patterns included: rim enhancing (n = 9), homogeneous (n = 7), nodule-in-nodule (n = 5), or heterogeneous (n = 3). A late peritumoral rim was observed in four (17%) of the hepatocellular carcinomas. Most metastases (17 of 18; 94%) demonstrated peripheral rim enhancement. The progressive centripetal enhancement of hemangiomas (n = 6) was clearly depicted. Focal nodular hyperplasia (n = 4) showed early homogeneous enhancement and one lesion demonstrated a central scar. CONCLUSION: The THRIVE-CENTRA-keyhole technique can be used to acquire single breath-hold, multiarterial images depicting improved enhancement characteristics of focal hepatic lesions. This technique will allow accurate timing of arterial scanning with 3D acquisition and high temporal resolution.