Using machine learning algorithms to review computed tomography scans and assess risk for cardiovascular disease: Retrospective analysis from the National Lung Screening Trial (NLST)

BACKGROUND

The National Lung Screening Trial (NLST) demonstrated that annual screening with low dose CT in high-risk population was associated with reduction in lung cancer mortality. Nonetheless, the leading cause of mortality in the study was from cardiovascular diseases.

PURPOSE

To determine whether the used machine learning automatic algorithms assessing coronary calcium score (CCS), level of liver steatosis and emphysema percentage in the lungs are good predictors of cardiovascular disease (CVD) mortality and incidence when applied on low dose CT scans.

MATERIALS AND METHODS

Three fully automated machine learning algorithms were used to assess CCS, level of liver steatosis and emphysema percentage in the lung. The algorithms were used on low-dose computed tomography scans acquired from 12,332 participants in NLST.

RESULTS

In a multivariate analysis, association between the three algorithm scores and CVD mortality have shown an OR of 1.72 (p = 0.003), 2.62 (p < 0.0001) for CCS scores of 101-400 and above 400 respectively, and an OR of 1.12 (p = 0.044) for level of liver steatosis. Similar results were shown for the incidence of CVD, OR of 1.96 (p < 0.0001), 4.94 (p < 0.0001) for CCS scores of 101-400 and above 400 respectively. Also, emphysema percentage demonstrated an OR of 0.89 (p < 0.0001). Similar results are shown for univariate analyses of the algorithms.

CONCLUSION

The three automated machine learning algorithms could help physicians to assess the incidence and risk of CVD mortality in this specific population. Application of these algorithms to existing LDCT scans can provide valuable health care information and assist in future research.

Automated opportunistic osteoporotic fracture risk assessment using computed tomography scans to aid in FRAX underutilization

Methods for identifying patients at high risk for osteoporotic fractures, including dual-energy X-ray absorptiometry (DXA)^1,2 and risk predictors like the Fracture Risk Assessment Tool (FRAX)^3-6, are underutilized. We assessed the feasibility of automatic, opportunistic fracture risk evaluation based on routine abdomen or chest computed tomography (CT) scans. A CT-based predictor was created using three automatically generated bone imaging biomarkers (vertebral compression fractures (VCFs), simulated DXA T-scores and lumbar trabecular density) and CT metadata of age and sex. A cohort of 48,227 individuals (51.8% women) aged 50-90 with available CTs before 2012 (index date) were assessed for 5-year fracture risk using FRAX with no bone mineral density (BMD) input (FRAXnb) and the CT-based predictor. Predictions were compared to outcomes of major osteoporotic fractures and hip fractures during 2012-2017 (follow-up period). Compared with FRAXnb, the major osteoporotic fracture CT-based predictor presented better receiver operating characteristic area under curve (AUC), sensitivity and positive predictive value (PPV) (+1.9%, +2.4% and +0.7%, respectively). The AUC, sensitivity and PPV measures of the hip fracture CT-based predictor were noninferior to FRAXnb at a noninferiority margin of 1%. When FRAXnb inputs are not available, the initial evaluation of fracture risk can be done completely automatically based on a single abdomen or chest CT, which is often available for screening candidates^7,8.

Simulating Dual-Energy X-Ray Absorptiometry in CT Using Deep-Learning Segmentation Cascade

PURPOSE

Osteoporosis is an underdiagnosed condition despite effective screening modalities. Dual-energy x-ray absorptiometry (DEXA) screening, although recommended in clinical guidelines, remains markedly underutilized. In contrast to DEXA, CT utilization is high and presents a valuable data source for opportunistic osteoporosis screening. The purpose of this study was to describe a method to simulate lumbar DEXA scores from routinely acquired CT studies using a machine-learning algorithm.

METHODS

Between January 2010 and September 2014, 610 CT studies of the abdomen and pelvis were used to develop spinal column and L1 to L4 multiclass segmentation. DEXA simulation training and validation used 1,843 pairs of CT studies accompanied by DEXA results obtained within a 6-month interval from the same individual. Machine learning-based regression was used to determine correlation between calculated grade (on the basis of vertebrae L1-L4) and DEXA t score.

RESULTS

Analysis of the t score equivalent, generated by the algorithm, revealed true positives in 1,144 patients, false positives in 92 patients, true negatives in 245 patients, and false negatives in 212 patients, resulting in an accuracy of 82%. Sensitivity for the detection of osteoporosis or osteopenia was 84.4% (95% confidence interval, 82.3%-86.2%), and specificity was 72.7% (95% confidence interval, 67.7%-77.2%).

CONCLUSIONS

The presented algorithm can identify osteoporosis and osteopenia with a high degree of accuracy (82%) and a small proportion of false positives. Efforts to cull greater information using machine-learning algorithms from pre-existing data have the potential to have a marked impact on population health efforts such as bone mineral density screening for osteoporosis, in which gaps in screening currently exist.