A Deep Learning Model for Automated Segmentation of Geographic Atrophy Imaged Using Swept-Source OCT

PURPOSE

To present a deep learning algorithm for segmentation of geographic atrophy (GA) using en face swept-source OCT (SS-OCT) images that is accurate and reproducible for the assessment of GA growth over time.

DESIGN

Retrospective review of images obtained as part of a prospective natural history study.

SUBJECTS

Patients with GA (n = 90), patients with early or intermediate age-related macular degeneration (n = 32), and healthy controls (n = 16).

METHODS

An automated algorithm using scan volume data to generate 3 image inputs characterizing the main OCT features of GA-hypertransmission in subretinal pigment epithelium (sub-RPE) slab, regions of RPE loss, and loss of retinal thickness-was trained using 126 images (93 with GA and 33 without GA, from the same number of eyes) using a fivefold cross-validation method and data augmentation techniques. It was tested in an independent set of one hundred eighty 6 × 6-mm² macular SS-OCT scans consisting of 3 repeated scans of 30 eyes with GA at baseline and follow-up as well as 45 images obtained from 42 eyes without GA.

MAIN OUTCOME MEASURES

The GA area, enlargement rate of GA area, square root of GA area, and square root of the enlargement rate of GA area measurements were calculated using the automated algorithm and compared with ground truth calculations performed by 2 manual graders. The repeatability of these measurements was determined using intraclass coefficients (ICCs).

RESULTS

There were no significant differences in the GA areas, enlargement rates of GA area, square roots of GA area, and square roots of the enlargement rates of GA area between the graders and the automated algorithm. The algorithm showed high repeatability, with ICCs of 0.99 and 0.94 for the GA area measurements and the enlargement rates of GA area, respectively. The repeatability limit for the GA area measurements made by grader 1, grader 2, and the automated algorithm was 0.28, 0.33, and 0.92 mm², respectively.

CONCLUSIONS

When compared with manual methods, this proposed deep learning-based automated algorithm for GA segmentation using en face SS-OCT images was able to accurately delineate GA and produce reproducible measurements of the enlargement rates of GA.

Using an Automated Algorithm to Identify Potential Drug-Induced Liver Injury Cases in a Pharmacovigilance Database

INTRODUCTION

Drug-induced liver injury (DILI) is the most frequent cause of acute liver failure in North America and Europe, but it is often missed because of unstandardized diagnostic methods and criteria. This study aimed to develop and validate an automated algorithm to identify potential DILI cases in routine pharmacovigilance (PV) activities.

METHODS

Post-marketing hepatic adverse events reported for a potentially hepatotoxic drug in a global PV database from 19 March 2017 to 18 June 2018 were assessed manually and with the automated algorithm. The algorithm provided case assessments by applying pre-specified criteria to all case data and narratives simultaneously.

RESULTS

A total of 1456 cases were included for analysis and assessed manually. Sufficient data for algorithm assessment were available for 476 cases (32.7%). Of these cases, manual assessment identified 312 (65.5%) potential DILI cases while algorithm assessment identified 305 (64.1%) potential DILI cases. Comparison of manual and algorithm assessments demonstrated a sensitivity of 97.8% and a specificity of 79.3% for the algorithm. Given the prevalence of potential DILI cases in the population studied, the algorithm was calculated to have positive predictive value 56.3% and negative predictive value 99.2%. The time required for manual review compared to algorithm review suggested that application of the algorithm prior to manual screening would have resulted in a time savings of 42.2%.

CONCLUSION

An automated algorithm to identify potential DILI cases was developed and successfully implemented. The algorithm demonstrated a high sensitivity, a high negative predictive value, along with significant efficiency and utility in a real-time PV database.

Reliability of foveal avascular zone metrics automatically measured by Cirrus optical coherence tomography angiography in healthy subjects

PURPOSE

To investigate the reliability of the foveal avascular zone (FAZ) metrics automatically measured using Cirrus optical coherence tomography angiography (OCTA) embedded algorithm compared to human manual measurement.

METHODS

Thirty-five eyes of 35 healthy subjects were enrolled and scanned four times continuously on Zeiss Cirrus HD-OCT 5000. The FAZ metrics (area, circularity and perimeter) of the superficial capillary plexus were measured automatically using the embedded tool and manually measured by the two independent observers using ImageJ. The repeatability of the four scans within all methods of measurements was calculated. The agreement of the manual vs automated measurement was also analyzed.

RESULTS

The repeatability of the automated algorithm was only poor to moderate (intraclass correlation coefficients [ICCs] for the area, perimeter and circularity were 0.600, 0.405 and 0.221, respectively) while the repeatability of the manually measured FAZ area and perimeter was good [([ICCs] ranged from 0.845 to 0.877) except the circularity (ICC = 0.538 to 0.608)]. The ranges of 95% limits of agreement between the manual measurements by the two observers were only 20% to 31% of those of automated-manual agreement. The Cirrus inbuilt algorithm obviously outlined the border of FAZ wrongly in 22.9% cases.

CONCLUSION

Caution should be taken when using the automated measurement results of FAZ metrics in Cirrus OCTA, because of the low repeatability and poor agreement compared with the manual measurement.

Comparative spatial resolution of 12-lead electrocardiography and an automated algorithm

BACKGROUND

The spatial resolution of pacemapping using 12-lead electrocardiography (ECG) or PaSo software is unknown.

OBJECTIVE

The purpose of this study was to determine the spatial resolution of traditional ECG pacemapping and pacemapping using the PaSo coefficients.

METHODS

Seventeen patients undergoing ablation of supraventricular tachycardias or atrioventricular node were included. After ablation, chamber (right ventricular outflow tract/rest of the right ventricle/left ventricle) geometry was created with Carto 3. Pacingwas performed from any point in these cardiac regions, the QRS morphology being the template and the point being considered as arrhythmia "origin." Subsequently, pacing was performed from points around the "origin" (1538 points). The QRS of these tagged points were compared by traditional ECG pacemapping and PaSo coefficients. The spatial resolution was calculated using correlations between the distance away from the origin (measured by 3 computational methods) and traditional ECG pacemapping and PaSo coefficients, independently.

RESULTS

A 0.01-unit decrease in the PaSo coefficient resulted in 1.1 mm increased Cartesian distance (95% confidence interval [CI] 0.9-1.3 mm; P < .001) and 2.4 mm increased geodesic distance (95% CI 1.9-2.9 mm; P < .001) and 664 mm³ increase in convex hull volume (95% CI 423-906 mm³; P < .0001). For traditional ECG pacemapping, each decrease in lead match resulted in 1.7 mm increased Cartesian distance (95% CI 1.5-2.0 mm; P < .001) and 3.4 mm increased geodesic distance (95% CI 2.8-4.1 mm; P < .001) and 712 mm³ increase in convex hull volume (95% CI 599-830 mm³; P < .0001). Both PaSo coefficients and traditional pacemapping showed a significant inverse linear correlation with distance from the "origin."

CONCLUSION

The resolution of mapping using the Paso software is better than that of traditional pacemapping.

Capturing fine-scale travel behaviors: a comparative analysis between personal activity location measurement system (PALMS) and travel diary

Background

Device-collected data from GPS and accelerometers for identifying active travel behaviors have dramatically changed research methods in transportation planning and public health. Automated algorithms have helped researchers to process large datasets with likely fewer errors than found in other collection methods (e.g., self-report travel diary). In this study, we compared travel modes identified by a commonly used automated algorithm (PALMS) that integrates GPS and accelerometer data with those obtained from travel diary estimates.

Methods

Sixty participants, who made 2100 trips during seven consecutive days of data collection, were selected from among the baseline sample of a project examining the travel behavior impact of a new light rail system in the greater Seattle, WA (USA) area. GPS point level analyses were first conducted to compare trip/place and travel mode detection results using contingency tables. Trip level analyses were then performed to investigate the effect of proportions of time overlap between travel logs and device-collected data on agreement rates. Global performance (with all subjects’ data combined) and subject-level performance of the algorithm were compared at the trip level.

Results

At the GPS point level, the overall agreement rate of travel mode detection was 77.4% between PALMS and the travel diary. The agreement rate for vehicular trip detection (84.5%) was higher than for bicycling (53.5%) and walking (58.2%). At the trip level, the global performance and subject-level performance of the PALMS algorithm were 46.4% and 42.4%, respectively. Vehicular trip detection showed highest agreement rates in all analyses. Study participants’ primary travel mode and car ownership were significantly related to the subject-level mode agreement rates.

Conclusions

The PALMS algorithm showed moderate identification power at the GPS point level. However, trip level analyses found lower agreement rates between PALMS and travel diary data, especially for active transportation. Testing different PALMS parameter settings may serve to improve the detection of active travel and help expand PALMS’s applicability in geographically different urbanized areas with a variety of travel modes.

Measurement of Left Ventricular Volumes and Ejection Fraction in Patients with Regional Wall Motion Abnormalities Using an Automated 3D Quantification Algorithm

Accurate and rapid left ventricular (LV) ejection fraction (EF) measurement is crucial for patients with wall motion abnormalities (WMAs). Conventional 2D echocardiographic imaging has limitations. The recently developed software HeartModel (HM, Philips Healthcare, Andover, MA, USA) has shown promise in automated 3D quantification. However, the accuracy and detailed features of HM in measurements of LV volume and EF in patients with regional WMAs have not been carefully investigated. In the present study, echocardiographic imaging (EPIQ, X5-1, Philips Healthcare) was performed in 72 patients with WMAs. The LV end-diastolic volume (EDV), end-systolic volume (ESV) and EF were measured by HM in three modes: without editing and with global and regional endocardial border editing (Auto 3D-NE, Auto 3D-GE and Auto 3D-RE, respectively). The conventional 2D Simpson's biplane method and manual 3D quantification (QLAB-3DQA software, Philips Healthcare), as the standard method, were used for comparison. Among the three HM modalities, Auto 3D-RE exhibited the best correlation with manual 3D in assessing EDV, ESV and EF (r = 0.88, 0.93 and 0.91, respectively), although it took slightly longer (67.3 ± 13.0 s). Auto 3D-RE also exhibited a small degree of bias for the measurements (EDV: 11.7mL, ESV: 8.45mL, EF: -1.57%) and narrow limits of agreement. Heterogeneity of LV wall motion was defined to indicate the dispersion degree of WMAs. It associated with the difference in EF measurement between Auto 3D-RE and manual 3D (p = 0.014, hazard ratio = 5.19). In patients with WMAs, HM with regional contour editing enables accurate and efficient evaluation of LV volume and EF.

Evaluation of an automated thresholding algorithm for the quantification of paraspinal muscle composition from MRI images

BACKGROUND

The imaging assessment of paraspinal muscle morphology and fatty infiltration has gained considerable attention in the past decades, with reports suggesting an association between muscle degenerative changes and low back pain (LBP). To date, qualitative and quantitative approaches have been used to assess paraspinal muscle composition. Though highly reliable, manual thresholding techniques are time consuming and not always feasible in a clinical setting. The tedious and rater-dependent nature of such manual thresholding techniques provides the impetus for the development of automated or semi-automated segmentation methods. The purpose of the present study was to develop and evaluate an automated thresholding algorithm for the assessment of paraspinal muscle composition. The reliability and validity of the muscle measurements using the new automated thresholding algorithm were investigated through repeated measurements and comparison with measurements from an established, highly reliable manual thresholding technique.

METHODS

Magnetic resonance images of 30 patients with LBP were randomly selected cohort of patients participating in a project on commonly diagnosed lumbar pathologies in patients attending spine surgeon clinics. A series of T2-weighted MR images were used to train the algorithm; preprocessing techniques including adaptive histogram equalization method image adjustment scheme were used to enhance the quality and contrast of the images. All muscle measurements were repeated twice using a manual thresholding technique and the novel automated thresholding algorithm, from axial T2-weigthed images, at least 5 days apart. The rater was blinded to all earlier measurements. Inter-method agreement and intra-rater reliability for each measurement method were assessed. The study did not received external funding and the authors have no disclosures.

RESULTS

There was excellent agreement between the two methods with inter-method reliability coefficients (intraclass correlation coefficients) varying from 0.79 to 0.99. Bland and Altman plots further confirmed the agreement between the two methods. Intra-rater reliability and standard error of measurements were comparable between methods, with reliability coefficient varying between 0.95 and 0.99 for the manual thresholding and 0.97-0.99 for the automated algorithm.

CONCLUSION

The proposed automated thresholding algorithm to assess paraspinal muscle size and composition measurements was highly reliable, with excellent agreement with the reference manual thresholding method.

Intratumoral heterogeneity as a source of discordance in breast cancer biomarker classification

BACKGROUND

Spatial heterogeneity in biomarker expression may impact breast cancer classification. The aims of this study were to estimate the frequency of spatial heterogeneity in biomarker expression within tumors, to identify technical and biological factors contributing to spatial heterogeneity, and to examine the impact of discordant biomarker status within tumors on clinical record agreement.

METHODS

Tissue microarrays (TMAs) were constructed using two to four cores (1.0 mm) for each of 1085 invasive breast cancers from the Carolina Breast Cancer Study, which is part of the AMBER Consortium. Immunohistochemical staining for estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) was quantified using automated digital imaging analysis. The biomarker status for each core and for each case was assigned using clinical thresholds. Cases with core-to-core biomarker discordance were manually reviewed to distinguish intratumoral biomarker heterogeneity from misclassification of biomarker status by the automated algorithm. The impact of core-to-core biomarker discordance on case-level agreement between TMAs and the clinical record was evaluated.

RESULTS

On the basis of automated analysis, discordant biomarker status between TMA cores occurred in 9 %, 16 %, and 18 % of cases for ER, PR, and HER2, respectively. Misclassification of benign epithelium and/or ductal carcinoma in situ as invasive carcinoma by the automated algorithm was implicated in discordance among cores. However, manual review of discordant cases confirmed spatial heterogeneity as a source of discordant biomarker status between cores in 2 %, 7 %, and 8 % of cases for ER, PR, and HER2, respectively. Overall, agreement between TMA and clinical record was high for ER (94 %), PR (89 %), and HER2 (88 %), but it was reduced in cases with core-to-core discordance (agreement 70 % for ER, 61 % for PR, and 57 % for HER2).

CONCLUSIONS

Intratumoral biomarker heterogeneity may impact breast cancer classification accuracy, with implications for clinical management. Both manually confirmed biomarker heterogeneity and misclassification of biomarker status by automated image analysis contribute to discordant biomarker status between TMA cores. Given that manually confirmed heterogeneity is uncommon (<10 % of cases), large studies are needed to study the impact of heterogeneous biomarker expression on breast cancer classification and outcomes.