Use of artificial intelligence in triaging of chest radiographs to reduce radiologists' workload

OBJECTIVES

To evaluate whether deep learning-based detection algorithms (DLD)-based triaging can reduce outpatient chest radiograph interpretation workload while maintaining noninferior sensitivity.

METHODS

This retrospective study included patients who underwent initial chest radiography at the outpatient clinic between June 1 and June 30, 2017. Readers interpreted radiographs with/without a commercially available DLD that detects nine radiologic findings (atelectasis, calcification, cardiomegaly, consolidation, fibrosis, nodules, pneumothorax, pleural effusion, and pneumoperitoneum). The reading order was determined in a randomized, crossover manner. The radiographs were classified into negative and positive examinations. In a 50% worklist reduction scenario, radiographs were sorted in descending order of probability scores: the lower half was regarded as negative exams, while the remaining were read with DLD by radiologists. The primary analysis evaluated noninferiority in sensitivity between radiologists reading all radiographs and simulating a 50% worklist reduction, with the inferiority margin of 5%. The specificities were compared using McNemar's test.

RESULTS

The study included 1964 patients (median age [interquartile range], 55 years [40-67 years]). The sensitivity was 82.6% (195 of 236; 95% CI: 77.5%, 87.3%) when readers interpreted all chest radiographs without DLD and 83.5% (197 of 236; 95% CI: 78.8%, 88.1%) in the 50% worklist reduction scenario. The difference in sensitivity was 0.8% (95% CI: - 3.8%, 5.5%), establishing noninferiority of 50% worklist reduction (p = 0.01). The specificity increased from 86.7% (1498 of 1728) to 90.4% (1562 of 1728) (p < 0.001) with DLD-based triage.

CONCLUSION

Deep learning-based triaging may substantially reduce workload without lowering sensitivity while improving specificity.

CLINICAL RELEVANCE STATEMENT

Substantial workload reduction without lowering sensitivity was feasible using deep learning-based triaging of outpatient chest radiograph; however, the legal responsibility for incorrect diagnoses based on AI-standalone interpretation remains an issue that should be defined before clinical implementation.

KEY POINTS

• A 50% workload reduction simulation using deep learning-based detection algorithm maintained noninferior sensitivity while improving specificity. • The CT recommendation rate significantly decreased in the disease-negative patients, whereas it slightly increased in the disease-positive group without statistical significance. • In the exploratory analysis, the noninferiority of sensitivity was maintained until 70% of the workload was reduced; the difference in sensitivity was 0%.

Case 322: Pyothorax-associated Lymphoma

HISTORY

An 81-year-old man living in South Korea was referred to the pulmonology clinic because of abnormal findings at routine surveillance CT. His past medical history included right radical nephroureterectomy for ureteral cancer in 2016, transurethral resection of a bladder tumor in 2015, and tuberculous pleurisy in his third decade of life that was complicated by a chronic calcified empyema. He had been doing well clinically until 6 months prior, when he presented to an outside hospital with progressive right-sided chest pain and dyspnea and was found to have active tuberculosis. During that hospitalization, he underwent chest CT and CT-guided biopsy of an incidentally found thoracic lesion, which revealed chronic active inflammation. His symptoms improved after initiation of antituberculous medication, and he was discharged home to complete treatment. Because of interval growth of this lesion noted on a subsequent surveillance CT scan, he was referred to pulmonology for further evaluation. Laboratory tests obtained during his visit revealed mild leukocytosis (1258 cells × 109/L; normal range, 4000-10 000 cells × 109/L) with neutrophilic predominance (82% neutrophils; normal range, 50%-75% neutrophils), and a mildly elevated C-reactive protein level (3.17 mg/dL; normal range, 0-0.5 mg/dL). A sputum culture was negative for tuberculosis. The patient reported mild chest discomfort and dyspnea. Liver MRI was performed to further evaluate an abnormal lesion seen at CT. Given the patient's recent nonspecific biopsy results and tuberculosis treatment, no further work-up was pursued. Contrast-enhanced chest CT was performed 6 months later because the patient developed worsening chest pain and dyspnea. He remained afebrile, with persistent leukocytosis (1485 cells × 109/L) and an elevated C-reactive protein level (3.56 mg/dL). On the basis of the imaging findings, repeat CT-guided biopsy and PET/CT were performed, thereby enabling confirmation of the diagnosis, and appropriate treatment was initiated.

Refined semi-lethal aerosol H5N1 influenza model in cynomolgus macaques for evaluation of medical countermeasures

Highly pathogenic avian influenza A H5N1 viruses cause high mortality in humans and have pandemic potential. Effective vaccines and treatments against this threat are urgently needed. Here, we have refined our previously established model of lethal H5N1 infection in cynomolgus macaques. An inhaled aerosol virus dose of 5.1 log10 plaque-forming unit (pfu) induced a strong febrile response and acute respiratory disease, with four out of six macaques succumbing after challenge. Vaccination with three doses of adjuvanted seasonal quadrivalent influenza vaccine elicited low but detectable neutralizing antibody to H5N1. All six vaccinated macaques survived four times the 50% lethal dose of aerosolized H5N1, while four of six unvaccinated controls succumbed to disease. Although vaccination did not protect against severe influenza, vaccinees had reduced respiratory dysfunction and lower viral load in airways compared to controls. We anticipate that our macaque model will play a vital role in evaluating vaccines and antivirals against influenza pandemics.

Case 322

HISTORY

An 81-year-old man living in South Korea was referred to the pulmonology clinic because of abnormal findings at routine surveillance CT (Fig 1A, 1B). His past medical history included right radical nephroureterectomy for ureteral cancer in 2016, transurethral resection of a bladder tumor in 2015, and tuberculous pleurisy in his third decade of life that was complicated by a chronic calcified empyema (Fig 1C). He had been doing well clinically until 6 months prior, when he presented to an outside hospital with progressive right-sided chest pain and dyspnea and was found to have active tuberculosis. During that hospitalization, he underwent chest CT and CT-guided biopsy of an incidentally found thoracic lesion (Fig 2), which revealed chronic active inflammation. His symptoms improved after initiation of antituberculous medication, and he was discharged home to complete treatment. Because of interval growth of this lesion noted on a subsequent surveillance CT scan, he was referred to pulmonology for further evaluation. Laboratory tests obtained during his visit revealed mild leukocytosis (1258 cells × 109/L; normal range, 4000-10 000 cells × 109/L) with neutrophilic predominance (82% neutrophils; normal range, 50%-75% neutrophils), and a mildly elevated C-reactive protein level (3.17 mg/dL; normal range, 0-0.5 mg/dL). A sputum culture was negative for tuberculosis. The patient reported mild chest discomfort and dyspnea. Liver MRI was performed to further evaluate an abnormal lesion seen at CT (Fig 3). Given the patient's recent nonspecific biopsy results and tuberculosis treatment, no further work-up was pursued. Contrast-enhanced chest CT was performed 6 months later because the patient developed worsening chest pain and dyspnea (Fig 4). He remained afebrile, with persistent leukocytosis (1485 cells × 109/L) and an elevated C-reactive protein level (3.56 mg/dL). On the basis of the imaging findings, repeat CT-guided biopsy and PET/CT were performed (Fig 5), thereby enabling confirmation of the diagnosis, and appropriate treatment was initiated.

Accuracy and complications of percutaneous transthoracic needle lung biopsy for the diagnosis of malignancy in patients with idiopathic pulmonary fibrosis

OBJECTIVES

To determine the accuracy of CT-guided percutaneous transthoracic needle lung biopsy (PTNB) for the diagnosis of malignancy and the associated complication rates in patients with idiopathic pulmonary fibrosis (IPF).

METHODS

This retrospective study included 91 CT-guided PTNBs performed in 80 patients with IPF from April 2003 through December 2016. Data regarding patients, target lesions, procedures, complications, and pathological reports were collected, and the final diagnosis was made. The diagnostic accuracy, sensitivity, specificity, percentage of nondiagnostic results, and complication rates were determined. Multivariable logistic regression analyses were performed to identify risk factors for nondiagnostic results and major complications.

RESULTS

Three biopsies (technical failure [n = 2] and undetermined final diagnosis [n = 1]) were excluded from the diagnostic accuracy calculation. The diagnostic accuracy, sensitivity, and specificity were 89% (78/88), 90% (62/69), and 84% (16/19), respectively. The percentage of nondiagnostic results was 34% (30/88). Lesion size ≤ 3 cm (odds ratio [OR], 8.8; 95% confidence interval [CI], 2.5-31.2; p = 0.001) and needle tip placement outside the target lesion (OR, 13.7; 95% CI, 1.4-132.2; p = 0.02) were risk factors for nondiagnostic results. The overall and major complication rates were 51% (46/91) and 12% (11/91), respectively. The presence of honeycombing along the path of the needle (OR, 11.2; 95% CI, 1.4-89.1; p = 0.02) was an independent risk factor for major complications.

CONCLUSIONS

CT-guided PTNB shows a relatively reasonable accuracy in diagnosing malignancy in patients with IPF. The complication rate may be high, especially when the needle passes through honeycomb lesions.

KEY POINTS

• In patients with idiopathic pulmonary fibrosis (IPF), CT-guided percutaneous transthoracic needle lung biopsy (PTNB) showed a relatively reasonable accuracy for the diagnosis of malignancy. • Target lesion size ≤ 3 cm and biopsy needle tip placement outside the target lesion were risk factors for nondiagnostic results of CT-guided PTNB. • The complication rate may be high, especially in cases where the biopsy needle passes through honeycomb lesions.

Diffuse pulmonary meningotheliomatosis

We report the case of an 80-year-old woman presenting with randomly distributed ground glass nodules in the lungs. Since this imaging appearance can be confusing and can mimic other disease processes, it is important to have an organized approach. In this specific case, the distribution and appearance of nodules, their presence for a prolonged period as well as the clinical context were clues to the diagnosis of diffuse pulmonary meningotheliomatosis (DPM). The final diagnosis was established by surgical biopsy. This article reviews the current literature on DPM, imaging appearance, and an algorithmic approach to the presented case.

Feasibility of diffusion and probabilistic white matter analysis in patients implanted with a deep brain stimulator

Deep brain stimulation (DBS) for Parkinson's disease (PD) is an established advanced therapy that produces therapeutic effects through high frequency stimulation. Although this therapeutic option leads to improved clinical outcomes, the mechanisms of the underlying efficacy of this treatment are not well understood. Therefore, investigation of DBS and its postoperative effects on brain architecture is of great interest. Diffusion weighted imaging (DWI) is an advanced imaging technique, which has the ability to estimate the structure of white matter fibers; however, clinical application of DWI after DBS implantation is challenging due to the strong susceptibility artifacts caused by implanted devices. This study aims to evaluate the feasibility of generating meaningful white matter reconstructions after DBS implantation; and to subsequently quantify the degree to which these tracts are affected by post-operative device-related artifacts. DWI was safely performed before and after implanting electrodes for DBS in 9 PD patients. Differences within each subject between pre- and post-implantation FA, MD, and RD values for 123 regions of interest (ROIs) were calculated. While differences were noted globally, they were larger in regions directly affected by the artifact. White matter tracts were generated from each ROI with probabilistic tractography, revealing significant differences in the reconstruction of several white matter structures after DBS. Tracts pertinent to PD, such as regions of the substantia nigra and nigrostriatal tracts, were largely unaffected. The aim of this study was to demonstrate the feasibility and clinical applicability of acquiring and processing DWI post-operatively in PD patients after DBS implantation. The presence of global differences provides an impetus for acquiring DWI shortly after implantation to establish a new baseline against which longitudinal changes in brain connectivity in DBS patients can be compared. Understanding that post-operative fiber tracking in patients is feasible on a clinically-relevant scale has significant implications for increasing our current understanding of the pathophysiology of movement disorders, and may provide insights into better defining the pathophysiology and therapeutic effects of DBS.

Reliability of automated versus handheld breast ultrasound examinations of suspicious breast masses

PURPOSE

The purpose of this study was to assess the reliability of automated breast ultrasound (ABUS) examinations of suspicious breast masses in comparison to handheld breast ultrasound (HHUS) with regard to Breast Imaging Reporting and Data System (BI-RADS) category assessment, and to investigate the factors affecting discrepancies in categorization.

METHODS

A total of 135 masses that were assessed as BI-RADS categories 4 and 5 on ABUS that underwent ultrasound (US)-guided core needle biopsy from May 2017 to December 2017 were included in this study. The BI-RADS categories were re-assessed using HHUS. Agreement of the BI-RADS categories was evaluated using kappa statistics, and the positive predictive value of each examination was calculated. Logistic regression analysis was performed to identify the mammography and US findings associated with discrepancies in the BI-RADS categorization.

RESULTS

The overall agreement between ABUS and HHUS in all cases was good (79.3%, kappa=0.61, P<0.001). Logistic regression analysis revealed that accompanying suspicious microcalcifications on mammography (odds ratio [OR], 4.63; 95% confidence interval [CI], 1.83 to 11.71; P=0.001) and an irregular shape on US (OR, 5.59; 95% CI, 1.43 to 21.83; P=0.013) were associated with discrepancies in the BI-RADS categorization.

CONCLUSION

The agreement between ABUS and HHUS examinations in the BI-RADS categorization of suspicious breast masses was good. The presence of suspicious microcalcifications on mammography and an irregular shape on US were factors associated with ABUS yielding a lower level of suspicion than HHUS in terms of the BI-RADS category assessment.

Medullary thyroid carcinoma: Application of Thyroid Imaging Reporting and Data System (TI-RADS) Classification

PURPOSE

To evaluate the applicability of ultrasound (US)-based Thyroid Imaging Reporting and Data System (TI-RADS) for evaluating medullary thyroid carcinoma (MTC).

MATERIALS AND METHODS

US images and medical records of patients with cytopathology-confirmed MTC between June 2003 and November 2016 were retrospectively reviewed. Four independent reviewers (two experienced and two inexperienced radiologists) evaluated 57 pre-operative US images of patients with MTC for shape, composition, echogenicity, margin, calcification of the MTC nodules, and categorized the nodules using TI-RADS classification. Weighted Kappa statistics was used to determine the inter-observer agreement of TI-RADS. Univariate and multivariate analyses were performed to assess US findings associated with lymph node metastasis.

RESULTS

Ninety-five percent of nodules were classified as either high suspicion (68%) or intermediate suspicion (26%). The overall inter-rater agreement was good (Kappa 0.84, agreement 91.52%), and inexperienced reviewers also showed good agreements with the most experienced reviewer (weighted Kappa 0.73 and 0.81). According to the univariate analysis, TI-RADS category 5, shape, microcalcification, and extrathyroid extension were significantly associated with lymph node metastasis in MTC patients (p = 0.003, 0.008, 0.001, and 0.021, respectively). As per the multivariate analysis, the presence of microcalcification and the irregular shape of the nodule were significantly associated with metastatic lymph nodes in MTC patients (odds ratio, 26.6; 95% CI, 2.7-263.7, p = 0.005, odds ratio, 14.7; 95% CI, 1.3-170, p = 0.031, respectively).

CONCLUSION

TI-RADS is applicable for the evaluation of MTC nodules with good inter-observer agreement.

Tumor heterogeneity of pancreas head cancer assessed by CT texture analysis: association with survival outcomes after curative resection

The value of image based texture features as a powerful method to predict prognosis and assist clinical management in cancer patients has been established recently. However, texture analysis using histograms and grey-level co-occurrence matrix in pancreas cancer patients has rarely been reported. We aimed to analyze the association of survival outcomes with texture features in pancreas head cancer patients. Eighty-eight pancreas head cancer patients who underwent preoperative CT images followed by curative resection were included. Texture features using different filter values were obtained. The texture features of average, contrast, correlation, and standard deviation with no filter, and fine to medium filter values as well as the presence of nodal metastasis were significantly different between the recurred (n = 70, 79.5%) and non-recurred group (n = 18, 20.5%). In the multivariate Cox regression analysis, lower standard deviation and contrast and higher correlation with lower average value representing homogenous texture were significantly associated with poorer DFS (disease free survival), along with the presence of lymph node metastasis. Texture parameters from routinely performed pre-operative CT images could be used as an independent imaging tool for predicting the prognosis in pancreas head cancer patients who underwent curative resection.

Impact of pharmaceutical excipients on in vitro association of saquinavir to chylomicrons

This study was performed to investigate the impact of pharmaceutical excipients commonly used for lymphatic transport on in vitro drug association with chylomicrons (CM). A CM association study was conducted using saquinavir solubilized in four different pharmaceutical excipients. We observed a linear relationship between saquinavir solubility and drug association, suggesting that the solubility of saquinavir in excipients is a key determinant for successful lymphatic delivery. Broadly, these results suggest that excipients with good solubilization properties may be advantageous for enhancing lymphatic drug delivery.

Phase II prevention trial of celecoxib in women at increased risk for breast cancer

1501

Background: Tamoxifen and raloxifen reduce the risk of estrogen receptor (ER) positive breast cancer. Acceptance of SERMs is low due to toxicities. New agents with a better toxicity profile which also will prevent ER-negative breast cancer are needed. The cyclooxygenase II inhibitor celecoxib has antiproliferative, and antiangiogenetic properties and is a putative breast cancer prevention agent. Our objective in this prospective short-term prevention study was to evaluate the effect of celecoxib on biomarkers in breast tissue and serum of high risk women. Methods: High risk individuals with a history of atypical hyperplasia, or lobular carcinoma insitu, or Gail risk greater than 1.67%, or previous history of ER negative breast cancer were eligible. After signing informed consent subjects underwent baseline fine needle aspiration (FNA) of the breast and serum collection for biomarker analysis. Subjects started celecoxib at 400 mg po BID for 6 months and underwent repeat FNA and serum collection. Biomarker endpoints included changes in ER and KI-67 expression in breast tissue analyzed by immunohistochemistry and insulin like growth factor binding protein (IGFBP-1) in serum analyzed by ELISA. The difference in biomarkers before and after treatment was assessed using a Wilcoxon signed rank test. Results: 49 patients were enrolled and accrual has been completed. Median age was 51 years and 57% of women were postmenopausal. The mean pre- and post treatment ER expression was 35.5% and 31%, respectively. The difference in ER expression was not significant (p = 0.1). The mean pre- and post treatment KI-67 expression was 1.68% and 1.60%, respectively. The difference was not significant (p = 0.8). However, there was a statistically significant difference in pre- and post treatment IGFBP-1 levels (p=0.023), with pre-and post treatment levels being 6.81ng/ml and 11.51ng/ml, respectively. No grade 3 or 4 toxicities were observed and there was no drop-out from the study. Conclusions: We have found a significant modulation of IGFBP-1 levels with 6 months celecoxib treatment in women who are at increased risk of developing breast cancer. The acceptable side effect profile and modulation of the biomarker (IGFBP-1) provides support for continued evaluation of celecoxib as a breast cancer prevention agent.

No significant financial relationships to disclose.

Turmeric (Curcuma longa) and curcumin inhibit the growth of Helicobacter pylori, a group 1 carcinogen

BACKGROUND

Curcumin, a polyphenolic chemical constituent derived from turmeric (Curcuma longa), has been shown to prevent gastric and colon cancers in rodents. Many mechanisms have been proposed for the chemopreventative effects, although the effect of curcumin on the growth of Helicobacter pylori has not been reported. H. pylori is a Group 1 carcinogen and is associated with the development of gastric and colon cancer.

MATERIALS AND METHODS

A methanol extract of the dried powdered turmeric rhizome and curcumin were tested against 19 strains of H. pylori, including 5 cagA+ strains.

RESULTS

Both the methanol extract and curcumin inhibited the growth of all strains of H. pylori in vitro with a minimum inhibitory concentration range of 6.25-50 micrograms/ml.

CONCLUSION

These data demonstrate that curcumin inhibits the growth of H. pylori cagA+ strains in vitro, and this may be one of the mechanisms by which curcumin exerts its chemopreventative effects.

Intravitreal pharmacokinetics of liposome-encapsulated amikacin in a rabbit model

BACKGROUND

Intravitreal injection of antibiotics has become a standard therapy for bacterial endophthalmitis. The duration of effective antimicrobial levels in the vitreous after single injection, however, may not be long enough to get optimal response. The authors prepared liposome-encapsulated amikacin for prolonging the duration of intravitreal therapeutic concentrations and investigated the intravitreal pharmacokinetics of the liposomes and amikacin in phosphate buffer solution (PBS) as control.

METHODS

The liposome-encapsulated amikacin was prepared by reverse-phase evaporation method. The intravitreal pharmacokinetics of the liposomes was compared with amikacin in PBS by fluorescence polarization immunoassay. Albino rabbits were randomly distributed into 12 groups. Rabbits in groups 1 to 6 and in groups I to VI (control groups) received an intravitreal injection of the liposome-encapsulated amikacin and amikacin in PBS, respectively.

RESULTS

The encapsulation rate of amikacin was 91%. The time of 50% spontaneous degradation (half-life) of the liposomes in PBS (38 degrees C, pH 7.4) was 47.6 days, and the time of 50% release (half-life) of the drug from the liposomes in PBS was 84.8 hours. The vitreous amikacin concentrations in groups 1 to 6 were significantly greater (P < 0.05) than those in control groups I to VI in every time interval, except in groups 1 to 3 at 1 hour after injection. The difference was particularly obvious in the endophthalmitis groups. The clearance of encapsulated amikacin in vitreous appeared to be related to the state of blood-ocular barrier and to the structural integrity of vitreous. The distribution, the absorption, and the elimination of encapsulated amikacin in vitreous showed the first-order kinetics.

CONCLUSION

The liposome-encapsulated amikacin prolonged half-life of the drug in vitreous. The results of the pharmacokinetic analysis suggested that in endophthalmitis, especially in severe cases, the liposomes may be preferable to conventional preparation.

[Causes of death and prognosis of myocardial infarction in patients with diabetes]

The long-term prognosis and causes of death of myocardial infarction (MI) in 62 patients with diabetes were studied. The mean follow-up time was 6.2 years. 11 cases died in the acute period of MI (8 weeks following onset of AMI), 9 cases of them (81.8%) had anterior infarction and their major causes of death were ventricular fibrillation and cardiogenic shock (72.2%). 19 cases died in the follow-up period, among them 14 cases (73.68%) had inferior and anterior-septal infarction; most of them died of reinfarction and sudden death. The cumulative survival rate 1.2 and 5 years after MI was 80.7%, 71.9% and 57.9%, respectively. The blood glucose level of the fatal group and the level of CPK and GOT of patients who died in the acute period were higher than those in the surviving group. The results suggest that treatment of myocardial infarction in patients with diabetes be more attentive to prevent ventricular fibrillation and cardiogenic shock during the acute period. After the acute period more attention should be paid to prevent reinfarction and and drop the blood glucose level at normal as possible.