Performance of an automated deep learning algorithm to identify hepatic steatosis within noncontrast computed tomography scans among people with and without HIV

PURPOSE

Hepatic steatosis (fatty liver disease) affects 25% of the world's population, particularly people with HIV (PWH). Pharmacoepidemiologic studies to identify medications associated with steatosis have not been conducted because methods to evaluate liver fat within digitized images have not been developed. We determined the accuracy of a deep learning algorithm (automatic liver attenuation region-of-interest-based measurement [ALARM]) to identify steatosis within clinically obtained noncontrast abdominal CT images compared to manual radiologist review and evaluated its performance by HIV status.

METHODS

We performed a cross-sectional study to evaluate the performance of ALARM within noncontrast abdominal CT images from a sample of patients with and without HIV in the US Veterans Health Administration. We evaluated the ability of ALARM to identify moderate-to-severe hepatic steatosis, defined by mean absolute liver attenuation <40 Hounsfield units (HU), compared to manual radiologist assessment.

RESULTS

Among 120 patients (51 PWH) who underwent noncontrast abdominal CT, moderate-to-severe hepatic steatosis was identified in 15 (12.5%) persons via ALARM and 12 (10%) by radiologist assessment. Percent agreement between ALARM and radiologist assessment of absolute liver attenuation <40 HU was 95.8%. Sensitivity, specificity, positive predictive value, and negative predictive value of ALARM were 91.7% (95%CI, 51.5%-99.8%), 96.3% (95%CI, 90.8%-99.0%), 73.3% (95%CI, 44.9%-92.2%), and 99.0% (95%CI, 94.8%-100%), respectively. No differences in performance were observed by HIV status.

CONCLUSIONS

ALARM demonstrated excellent accuracy for moderate-to-severe hepatic steatosis regardless of HIV status. Application of ALARM to radiographic repositories could facilitate real-world studies to evaluate medications associated with steatosis and assess differences by HIV status.

Accuracy of Resident-Performed Point-of-Care Lung Ultrasound Examinations Versus Chest Radiography in Pneumothorax Follow-up After Tube Thoracostomy in Rwanda

OBJECTIVES

The aim of this study was to evaluate the accuracy and timeliness of resident-performed point-of-care lung ultrasound (LUS) examinations for the follow-up of pneumothorax (PTX) after tube thoracostomy.

METHODS

After brief training, Rwandan surgical residents blinded to chest radiography (CXR) performed and interpreted LUS examinations for PTX in participants undergoing CXR for PTX follow-up. Treating clinicians interpreted CXR for the presence of PTX for therapeutic decisions. Lung ultrasound was later reviewed by ultrasound experts, and CXR was reviewed by a radiologist. We defined expert LUS interpretation as the reference standard. The sensitivity and specificity of resident-performed LUS examinations for diagnosing PTX were calculated. We assessed agreement between trained resident versus expert LUS and clinician versus radiology CXR using the Cohen κ coefficient. We compared the time to results between LUS and CXR.

RESULTS

Over an 8-month period, 51 participants were enrolled. Compared to expert LUS interpretation, the sensitivity and specificity (95% confidence intervals) of resident LUS were 100% (85%-100%) and 96% (82%-100%), respectively, whereas the sensitivity and specificity of clinician-interpreted CXR were 48% (27%-69%) and 100% (88%-100%). The agreement between resident and expert LUS was excellent (κ = 0.96), whereas the agreement between clinician and radiologist CXR was only moderate (κ = 0.60). The time to results was significantly longer for CXR than LUS (mean, 1335 versus 396 minutes; P = .0001).

CONCLUSIONS

A resident-performed LUS examination was a quicker imaging modality with superior sensitivity compared to clinician-interpreted CXR for PTX follow-up after tube thoracostomy in this Rwandan study. Lung ultrasound can be a valuable imaging tool for PTX follow-up, especially in resource-limited settings.

Colorectal Cancer:: Imaging Conundrums

Progressive technological advancements in imaging have significantly improved the preoperative sensitivity for the detection of very small foci of regionally- or hematogenously-metastatic colorectal cancer. Unfortunately, this information has not translated to continued linear gains in patient survival, and might even result in the false-positive upstaging of some cases: these are two conundrums in the imaging of colorectal cancer. Both conundrums might be resolved by the widespread use of real-time imaging guidance during operative procedures. This might open the way for the widespread use of fluorodeoxyglucose PET/CT for the initial staging of patients with colorectal cancer.

Improving Abnormality Detection on Chest Radiography Using Game-Like Reinforcement Mechanics

RATIONALE AND OBJECTIVES

Despite their increasing prevalence, online textbooks, question banks, and digital references focus primarily on explicit knowledge. Implicit skills such as abnormality detection require repeated practice on clinical service and have few digital substitutes. Using mechanics traditionally deployed in video games such as clearly defined goals, rapid-fire levels, and narrow time constraints may be an effective way to teach implicit skills.

MATERIALS AND METHODS

We created a freely available, online module to evaluate the ability of individuals to differentiate between normal and abnormal chest radiographs by implementing mechanics, including instantaneous feedback, rapid-fire cases, and 15-second timers. Volunteer subjects completed the modules and were separated based on formal experience with chest radiography. Performance between training and testing sets were measured for each group, and a survey was administered after each session.

RESULTS

The module contained 74 cases and took approximately 20 minutes to complete. Thirty-two cases were normal radiographs and 56 cases were abnormal. Of the 60 volunteers recruited, 25 were "never trained" and 35 were "previously trained." "Never trained" users scored 21.9 out of 37 during training and 24.0 out of 37 during testing (59.1% vs 64.9%, P value <.001). "Previously trained" users scored 28.0 out of 37 during training and 28.3 out of 37 during testing phases (75.6% vs 76.4%, P value = .56). Survey results showed that 87% of all subjects agreed the module is an efficient way of learning, and 83% agreed the rapid-fire module is valuable for medical students.

CONCLUSIONS

A gamified online module may improve the abnormality detection rates of novice interpreters of chest radiography, although experienced interpreters are less likely to derive similar benefits. Users reviewed the educational module favorably.

Subchondral insufficiency fracture of the knee: A recognizable associated soft tissue edema pattern and a similar distribution among men and women

OBJECTIVE

Primary: to describe the presence and pattern of soft tissue edema in subchondral insufficiency fractures of the knee (SIFK). Secondary: to investigate the gender distribution and identify factors associated with disease progression.

METHODS

MR images of 74 SIFKs in 74 patients were retrospectively reviewed for soft tissue edema presence and location, meniscal tears and extrusion and synovitis. The clinical records were reviewed for age, gender, and BMI. Follow up examinations were reviewed to assess for progression. Data were analyzed for gender distribution and for association between each imaging finding as a predictor of SIFK location and progression.

RESULTS

Soft tissue edema was present in 89% (66/74) of SIFK. It was located around the MCL in 78% (58/74), posterior to and abutting on the posterior distal femur in 68% (50/74), around to the tibia in only 18% (13/74), but when present it strongly predicted the presence of a medial tibial plateau SIFK (p=5.6×10^-12). Edema extended to the vastus medialis fascia in 51% (38/74) and vastus lateralis fascia in 24% (18/74). Gender distribution was 1:1 (males=38, females=36), most common in the 6th decade (29/74, 39%). Lesion progression showed a trend towards being more common in females (8/9, 89%) compared to males (9/16, 56%), and in patients with meniscal extrusion (≥3mm) (13/14, 93%) compared to those with no extrusion (2/7, 29%).

CONCLUSION

A recognizable soft tissue edema pattern is seen in SIFK and may have an important role in early diagnosis. Also, SIFK may affect equally males and females in the 6th decade and may progress more in females. Meniscal extrusion may predispose to disease progression.

Utilization and reporting of bone densitometry: what can the musculoskeletal radiologist do to help, rather than to hurt?

Osteoporosis is a highly prevalent disease that predisposes patients to fragility fractures. These fractures carry serious risks, including increased mortality and the potential loss of functional independence. Effective treatments for osteoporosis are available, but these should be initiated before a fragility fracture actually occurs; to do so, osteoporosis must be diagnosed while it is still asymptomatic. The gold standard screening test used to detect low bone mass is dual-energy x-ray absorptiometry (DXA). Despite its clinical importance, the DXA report is sometimes neglected by radiologists-as though it were somehow less significant in diagnosis than our other modalities. If musculoskeletal radiologists are to help, rather than to hurt, we must raise the profile of this critical test with evidence-based utilization and coherent reporting: detailed recommendations for doing so are available from professional organizations such as the International Society for Clinical Densitometry and the National Osteoporosis Foundation. This brief survey will seek to remind the radiologist that a good densitometry report requires more than just copying numbers from a scanner.

Increased error rates in preliminary reports issued by radiology residents working more than 10 consecutive hours overnight

RATIONALE AND OBJECTIVES

To determine if the rate of major discrepancies between resident preliminary reports and faculty final reports increases during the final hours of consecutive 12-hour overnight call shifts.

MATERIALS AND METHODS

Institutional review board exemption status was obtained for this study. All overnight radiology reports interpreted by residents on-call between January 2010 and June 2010 were reviewed by board-certified faculty and categorized as major discrepancies if they contained a change in interpretation with the potential to impact patient management or outcome. Initial determination of a major discrepancy was at the discretion of individual faculty radiologists based on this general definition. Studies categorized as major discrepancies were secondarily reviewed by the residency program director (M.H.S.) to ensure consistent application of the major discrepancy designation. Multiple variables associated with each report were collected and analyzed, including the time of preliminary interpretation, time into shift study was interpreted, volume of studies interpreted during each shift, day of the week, patient location (inpatient or emergency department), block of shift (2-hour blocks for 12-hour shifts), imaging modality, patient age and gender, resident identification, and faculty identification. Univariate risk factor analysis was performed to determine the optimal data format of each variable (ie, continuous versus categorical). A multivariate logistic regression model was then constructed to account for confounding between variables and identify independent risk factors for major discrepancies.

RESULTS

We analyzed 8062 preliminary resident reports with 79 major discrepancies (1.0%). There was a statistically significant increase in major discrepancy rate during the final 2 hours of consecutive 12-hour call shifts. Multivariate analysis confirmed that interpretation during the last 2 hours of 12-hour call shifts (odds ratio (OR) 1.94, 95% confidence interval (CI) 1.18-3.21), cross-sectional imaging modality (OR 5.38, 95% CI 3.22-8.98), and inpatient location (OR 1.81, 95% CI 1.02-3.20) were independent risk factors for major discrepancy.

CONCLUSIONS

In a single academic medical center, major discrepancies in resident preliminary reports increased significantly during the final 2 hours of consecutive 12-hour overnight call shifts. This finding could be related to either fatigue or circadian desynchronization. Discrimination of these two potential etiologies requires additional investigation as major discrepancies in resident reports have the potential to negatively impact patient care/outcome. Cross-sectional imaging modalities including computed tomography and ultrasound (versus conventional radiography), as well as inpatient location (versus Emergency Department location), were also associated with significantly higher major discrepancy rates.

Rehabilitation of pediatric spinal cord injury: From acute medical care to rehabilitation and beyond

OBJECTIVE

Spinal Cord Injury (SCI) in the pediatric population is an uncommon but potentially devastating entity that may be unfamiliar to many practitioners. Because its treatment requires a long-term, well-planned, integrated approach, understanding the natural history, treatment, and common comorbidities of pediatric SCI is crucial for all physicians caring for these children. We present a review of the current literature on this topic to discuss the current standard of medical and rehabilitation care of pediatric SCI patients.

DATA SOURCES

Studies for this review article were obtained by a thorough PubMed search, including but not limited to the terms "spinal cord injury," "SCI," "spinal rehabilitation," and "pediatric spinal injury." Frequently referenced articles of particular prominence in the field were also reviewed.

STUDY SELECTION

Studies were included primarily to illustrate specific clinical situations in this overview article.

DATA EXTRACTION

Studies were reviewed by the authors (JG and HK) and clinically important details were incorporated into the review.

DATA SYNTHESIS

Not applicable.

CONCLUSIONS

Pediatric SCI, while relatively uncommon, is a life-changing event for affected children and their families, physically, medically, and economically. Providing multidisciplinary care, as well as early and extensive rehabilitation, translates into superior outcomes. However, more research remains necessary to identify tactics for better community integration and increased educational, financial, and social success.

Chemoembolization and bland embolization of neuroendocrine tumor metastases to the liver

PURPOSE

To assess the toxicity and efficacy of chemoembolization and bland embolization in patients with neuroendocrine tumor metastases to the liver.

MATERIALS AND METHODS

A total of 67 patients underwent 219 embolization procedures: 23 patients received primarily bland embolization with PVA with or without iodized oil and 44 primarily received chemoembolization with cisplatin, doxorubicin, mitomycin-C, iodized oil, and polyvinyl alcohol. Clinical, laboratory, and imaging follow-up was performed 1 month after completion of therapy and every 3 months thereafter. Patients with disease relapse were treated again when feasible. Toxicity was assessed according to National Cancer Institute Common Toxicity Criteria for Adverse Events, version 3.0. Efficacy was assessed by clinical and morphologic response. Time to progression (TTP), time to treatment failure, and survival were estimated by Kaplan-Meier analysis.

RESULTS

Ten of 67 patients (15%) were lost to follow-up. The mortality rate at 30 days was 1.4%. Toxicities of grade 3 or worse in severity occurred after 25% of chemoembolization procedures and 22% of bland embolization procedures (odds ratio, 1.2; 95% CI, 0.4-4.0). Mean length of stay was 1.5 day in both groups. Rates of freedom from progression at 1, 2, and 3 years were 49%, 49%, and 35% after chemoembolization and 0%, 0%, and 0% after bland embolization (log-rank test, P = .16). Among the subgroup with carcinoid tumors, the proportions without progression were 65%, 65%, and 52% after chemoembolization and 0%, 0%, and 0% after bland embolization (log-rank test, P = .08). Patients treated with chemoembolization and bland embolization experienced symptomatic relief for means of 15 and 7.5 months, respectively (P = .14). Survival rates at 1, 3, and 5 years after therapy were 86%, 67%, and 50%, respectively, after chemoembolization and 68%, 46%, and 33%, respectively, after bland embolization (log-rank test, P = .18).

CONCLUSIONS

Chemoembolization was not associated with a higher degree of toxicity than bland embolization. Chemoembolization demonstrated trends toward improvement in TTP, symptom control, and survival. Based on these results, a multicenter prospective randomized trial is warranted.