Strong versus Weak Data Labeling for Artificial Intelligence Algorithms in the Measurement of Geographic Atrophy.

Purpose: To gain an understanding of data labeling requirements to train deep learning models for measurement of geographic atrophy (GA) with fundus autofluorescence (FAF) images. Design: Evaluation of artificial intelligence (AI) algorithms. Subjects: The Age-Related Eye Disease Study 2 (AREDS2) images were used for training and cross-validation, and GA clinical trial images were used for testing. Methods: Training data consisted of 2 sets of FAF images; 1 with area measurements only and no indication of GA location (Weakly labeled) and the second with GA segmentation masks (Strongly labeled). Main Outcome Measures: Bland-Altman plots and scatter plots were used to compare GA area measurement between ground truth and AI measurements. The Dice coefficient was used to compare accuracy of segmentation of the Strong model. Results: In the cross-validation AREDS2 data set (n = 601), the mean (standard deviation [SD]) area of GA measured by human grader, Weakly labeled AI model, and Strongly labeled AI model was 6.65 (6.3) mm2, 6.83 (6.29) mm2, and 6.58 (6.24) mm2, respectively. The mean difference between ground truth and AI was 0.18 mm2 (95% confidence interval, [CI], -7.57 to 7.92) for the Weakly labeled model and -0.07 mm2 (95% CI, -1.61 to 1.47) for the Strongly labeled model. With GlaxoSmithKline testing data (n = 156), the mean (SD) GA area was 9.79 (5.6) mm2, 8.82 (4.61) mm2, and 9.55 (5.66) mm2 for human grader, Strongly labeled AI model, and Weakly labeled AI model, respectively. The mean difference between ground truth and AI for the 2 models was -0.97 mm2 (95% CI, -4.36 to 2.41) and -0.24 mm2 (95% CI, -4.98 to 4.49), respectively. The Dice coefficient was 0.99 for intergrader agreement, 0.89 for the cross-validation data, and 0.92 for the testing data. Conclusions: Deep learning models can achieve reasonable accuracy even with Weakly labeled data. Training methods that integrate large volumes of Weakly labeled images with small number of Strongly labeled images offer a promising solution to overcome the burden of cost and time for data labeling. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Implementation of a Large-Scale Image Curation Workflow Using Deep Learning Framework.

Purpose: The curation of images using human resources is time intensive but an essential step for developing artificial intelligence (AI) algorithms. Our goal was to develop and implement an AI algorithm for image curation in a high-volume setting. We also explored AI tools that will assist in deploying a tiered approach, in which the AI model labels images and flags potential mislabels for human review. Design: Implementation of an AI algorithm. Participants: Seven-field stereoscopic images from multiple clinical trials. Methods: The 7-field stereoscopic image protocol includes 7 pairs of images from various parts of the central retina along with images of the anterior part of the eye. All images were labeled for field number by reading center graders. The model output included classification of the retinal images into 8 field numbers. Probability scores (0-1) were generated to identify misclassified images, with 1 indicating a high probability of a correct label. Main Outcome Measures: Agreement of AI prediction with grader classification of field number and the use of probability scores to identify mislabeled images. Results: The AI model was trained and validated on 17 529 images and tested on 3004 images. The pooled agreement of field numbers between grader classification and the AI model was 88.3% (kappa, 0.87). The pooled mean probability score was 0.97 (standard deviation [SD], 0.08) for images for which the graders agreed with the AI-generated labels and 0.77 (SD, 0.19) for images for which the graders disagreed with the AI-generated labels (P < 0.0001). Using receiver operating characteristic curves, a probability score of 0.99 was identified as a cutoff for distinguishing mislabeled images. A tiered workflow using a probability score of < 0.99 as a cutoff would include 27.6% of the 3004 images for human review and reduce the error rate from 11.7% to 1.5%. Conclusions: The implementation of AI algorithms requires measures in addition to model validation. Tools to flag potential errors in the labels generated by AI models will reduce inaccuracies, increase trust in the system, and provide data for continuous model development.

Impact of the National Lung Screening Trial (NLST) publication and Medicare Lung cancer screening payment on lung cancer incidence rates: An interrupted time series analysis.

INTRODUCTION: Explore the impact of the Lung Cancer Screening Trial (NLST-September-2011) and the Medicare approval for CT-screening (CT-LCS-AP-February-2015) on lung cancer incidence rates, mortality, and the percentage of early-stage lung cancer diagnosis (ESLCD-T1-T2N0M0). METHODS: Retrospective interrupted time series analysis using SEER-18 database. All individuals with lung cancer (LC) diagnosis from 2006 to 2016 were included. The effect of NLST and CT-AP-2015 on the monthly percentage of early-stage ESLCD was the primary outcome, additionally LC incidence and mortality rates were calculated. The analysis was performed by age, sex, race, marital status, insurance status, and household income. Bivariate and multivariate models were used to identify predictors of ESLCD. RESULTS: The study cohort was composed by 388,207 individuals, 69 years old in average, 46.6 % female, and 81.1 % white. LC incidence and mortality rates declined from 2006 to 2016 without association with NLST-September-2011 and CT-LCS-AP-February-2015. The percentage of ESLCD increased over time for all groups. Overall rates of ESLCD started at 18 % in January-2006 and increased to 25 % by December-2016. The intervention NLST-2011 did not show an impact in the ESLCD while the CT-AP-2015 showed a significant impact in the ESLCD trend (p < 0.001). ESLCD was associated with female, white, insurance, and household incomes above median. Medicare expansion was a significant factor for insured group, married patients and those from households under the median income level. CONCLUSION: Medicare approval for CT screening was found to have a statistically significant effect on the diagnosis of early-stage lung cancer and neither NLST-September-2011 nor CT-AP-2015-February-2015 impacted the incidence nor mortality rates. POLICY SUMMARY: To improve early-stage lung cancer diagnosis, it is vital to invest in health policies to increase Lung Cancer Screening implementation and to reduce disparities in access to diagnosis. Furthermore, policies that facilitate access to diagnosis and treatment are crucial to reduce lung cancer mortality.

Redefining the Spectrum of Pentosan Polysulfate Retinopathy: Multimodal Imaging Findings from a Cross-Sectional Screening Study.

PURPOSE: There is growing evidence of a direct association between pentosan polysulfate (PPS) therapy and the development of macular changes. Using standardized visual acuity (VA) testing and multimodal imaging, we investigated the impact of PPS therapy on vision and described an expanded spectrum of imaging findings among PPS users. DESIGN: Cross-sectional screening study. PARTICIPANTS: Thirty-nine patients who were current or recent users of PPS. METHODS: The participants underwent a brief eye examination and answered a comprehensive medical and ophthalmic history questionnaire. Color fundus photography, fundus autofluorescence (FAF), and spectral-domain OCT (SD-OCT) were performed. The images were evaluated by expert graders at Wisconsin Reading Center. Abnormalities were categorized as definite toxicity (DT) if seen on both FAF and SD-OCT and as questionable toxicity (QT) if seen on either FAF or SD-OCT. MAIN OUTCOME MEASURES: ETDRS and Snellen VA, the dosage and duration of PPS exposure, and the prevalence of retinal toxicity on imaging. RESULTS: The mean ETDRS and Snellen VA of the study cohort were 85 letters and 20/22, respectively. The mean PPS daily dose was 282 mg (range, 88-400 mg), whereas the mean cumulative dose was 915 g (range, 19-3650 g) over a mean period of 8.8 years (range, 2 months-25 years). There was evidence of retinopathy in 41% of the eyes; DT was identified in 24 eyes (31%) and QT in 8 eyes (10%). Retinal pigment epithelium (RPE) abnormalities (thickening or thinning or both) were present in all eyes with DT. Retinal pigment epithelium atrophy was seen in 7 eyes (9%). In addition to well-established findings, the unique SD-OCT features of this cohort included interdigitation zone abnormalities and the presence of a flying saucer-type defect. Fundus autofluorescence abnormalities were seen in 24 eyes (30.8%), with 20 (66.7%) of these exhibiting abnormalities located outside the central subfield and extending beyond the arcades. CONCLUSIONS: Findings from the masked grading of multimodal imaging at a centralized reading center suggest a wider phenotypic spectrum of structural abnormalities among patients taking PPS. Macular changes selectively involve the RPE and outer retina, with a range of findings often seen beyond the arcades. The subtle and atypical findings in this cohort should prompt clinicians to consider lowering the threshold for diagnosing PPS retinopathy.

Traumatic Brain Injury Screening and the Unmet Health Needs of Shelter-Seeking Women with Head Injuries Related to Intimate Partner Violence.

Background: Unmet health needs of women with head injuries sustained by intimate partner violence (IPV) include risk of traumatic brain injury (TBI). The purpose of this evaluation was to explore the potential effectiveness of TBI screening as a health promotion strategy for shelter-seeking women with IPV head injuries. We wanted to learn if shelter-seeking women, willing to disclose IPV, would accept TBI screening if offered. Methods: An extended version of the HELPS TBI screening tool and survey of daily symptoms and health needs were used to screen new residents of an urban shelter for women. Results: The participants (N = 18) primarily were educated black women with one or more self-reported IPV-related head injury. Most participants (77.8%) had positive TBI screens for probable brain injury. The majority (88.8%) lived with one or more daily symptoms they did not have before sustaining a IPV head injury. The symptoms reported most frequently were depression (88.9%), anxiety (77.8%), and headache (66.7%). All participants had one or more unmet health need. Although most (77.8%) needed to see a primary care provider, mental health care was the most important health need identified. Conclusions: TBI screening could be considered an effective health promotion strategy for IPV survivors if screening facilitates treatment for positive screens and other unmet health needs. Further research is needed to properly assess this.

The Incidence of Node-Positive Non-small-Cell Lung Cancer Undergoing Sublobar Resection and the Role of Radiation in Its Management.

Purpose: To identify the incidence, preoperative risk factors, and prognosis associated with pathologically positive lymph node (pN+) in patients undergoing a sub-lobar resection (SLR). Methods: This is a retrospective study using the National Cancer Database (NCDB) from 2004 to 2014 analyzing SLR excluding those with any preoperative chemotherapy and/or radiation, follow-up <3 months, stage IV disease, or >1 tumor nodule. Multivariable modeling (MVA) was used to determine factors associated with overall survival (OS). Propensity score matching (PSM) was used to determine preoperative risk factors for pN+ in patients having at least one node examined to assess radiation's effect on OS in those patients with pN+ and to determine whether SLR was associated with inferior OS as compared to lobectomy for each nodal stage. Results: A total of 40,202 patients underwent SLR, but only 58.3% had one lymph node examined. Then, 2,615 individuals had pN+ which decreased progressively from 15.1% in 2004 to 8.9% in 2014 (N1, from 6.3 to 3.0%, and N2, from 8.4 to 5.9%). A lower risk of pN+ was noted for squamous cell carcinomas, bronchioloalveolar adenocarcinoma (BAC), adenocarcinomas, and right upper lobe locations. In the pN+ group, OS was worse without chemotherapy or radiation. Radiation was associated with a strong trend for OS in the entire pN+ group (p = 0.0647) which was largely due to the effects on those having N2 disease (p = 0.009) or R1 resections (p = 0.03), but not N1 involvement (p = 0.87). PSM noted that SLR was associated with an inferior OS as compared to lobectomy by nodal stage in the overall patient population and even for those with tumors <2 cm. Conclusion: pN+ incidence in SLRs has decreased over time. SLR was associated with inferior OS as compared to lobectomy by nodal stage. Radiation appears to improve the OS in patients undergoing SLR with pN+, especially in those with N2 nodal involvement and/or positive margins.

Diabetes mellitus type 2 is associated with increased tumor expression of programmed death-ligand 1 (PD-L1) in surgically resected non-small cell lung cancer-A matched case-control study.

OBJECTIVES: Programmed death-ligand 1 (PD-L1) expression is a biomarker for cancer immunotherapy. Diabetes mellitus type-2 is a comorbid disease associated with adverse outcomes in Non-Small Cell Lung Cancer (NSCLC). We aimed to investigate the differences in PD-L1 expression in diabetics. METHODS: A matched case-control cohort of surgically-resected NSCLC was assembled from an early multicenter study (PMID: 19152440). PD-L1 immunohistochemistry (Clone 22C3) was graded by a tumor positive score (TPS) system (TPS0: no staining; TPS1: <1%; TPS2: 1-49%; TPS3: ≥50%). Variables showing significance at univariate survival analysis were fit in a Cox regression survival model. RESULTS: Diabetics (n=40) and nondiabetics (n=39) showed no differences in age, gender, cancer stage, and follow-up. NSCLCs were more likely PD-L1 positive in diabetics but with tumor positivity <50% (TPS0: 7.5 vs. 20.5%, TPS1: 35 vs. 25.6%, TPS2: 45 vs.23.1%, TPS3: 12.5 vs. 30.8%, respectively; P<0.05). In diabetics, squamous cell carcinomas (SCC) and adenocarcinomas were mainly TPS2 (65% vs. 20%) and TPS1 (50% vs. 26%), respectively. Peritumoral inflammation correlated with TPS (r=0.228), a relationship accentuated in diabetics (r=0.377, P<0.05) but diminished and non-significant in nondiabetics (r=0.136, P≥0.05). This association was stronger in SCC (r=0.424). Diabetes was associated with increased tumor recurrence (HR: 3.08; 95%CI: 1.027-9.23). CONCLUSION: Diabetes is associated with an increase in peritumoral inflammation, PD-L1 positivity, and recurrence in NSCLC, more pronounced in SCC, suggesting the possibility of metabolic reprogramming and upregulation of PD-L1 by inducible pathways.

Vitamin D status and early age-related macular degeneration in postmenopausal women.

OBJECTIVE: The relationship between serum 25-hydroxyvitamin D (25[OH]D) concentrations (nmol/L) and the prevalence of early age-related macular degeneration (AMD) was investigated in participants of the Carotenoids in Age-Related Eye Disease Study. METHODS: Stereoscopic fundus photographs, taken from 2001 to 2004, assessed AMD status. Baseline (1994-1998) serum samples were available for 25(OH)D assays in 1313 women with complete ocular and risk factor data. Odds ratios (ORs) and 95% confidence intervals (CIs) for early AMD (n = 241) of 1287 without advanced disease were estimated with logistic regression and adjusted for age, smoking, iris pigmentation, family history of AMD, cardiovascular disease, diabetes, and hormone therapy use. RESULTS: In multivariate models, no significant relationship was observed between early AMD and 25(OH)D (OR for quintile 5 vs 1, 0.79; 95% CI, 0.50-1.24; P for trend = .47). A significant age interaction (P = .002) suggested selective mortality bias in women aged 75 years and older: serum 25(OH)D was associated with decreased odds of early AMD in women younger than 75 years (n = 968) and increased odds in women aged 75 years or older (n = 319) (OR for quintile 5 vs 1, 0.52; 95% CI, 0.29-0.91; P for trend = .02 and OR, 1.76; 95% CI, 0.77-4.13; P for trend = .05, respectively). Further adjustment for body mass index and recreational physical activity, predictors of 25(OH)D, attenuated the observed association in women younger than 75 years. Additionally, among women younger than 75 years, intake of vitamin D from foods and supplements was related to decreased odds of early AMD in multivariate models; no relationship was observed with self-reported time spent in direct sunlight. CONCLUSIONS: High serum 25(OH)D concentrations may protect against early AMD in women younger than 75 years.

Healthy lifestyles related to subsequent prevalence of age-related macular degeneration.

OBJECTIVE: To investigate the relationships between lifestyle behaviors of diet, smoking, and physical activity and the subsequent prevalence of age-related macular degeneration (AMD). METHODS: The population included 1313 participants (aged 55-74 years) in the Carotenoids in Age-Related Eye Disease Study, an ancillary study of the Women's Health Initiative Observational Study. Scores on a modified 2005 Healthy Eating Index were assigned using responses to a food frequency questionnaire administered at baseline of the Women's Health Initiative Observational Study (1994-1998). Physical activity and lifetime smoking history were queried. An average of 6 years later, stereoscopic fundus photographs were taken to assess the presence and severity of AMD; it was present in 202 women, 94% of whom had early AMD, the primary outcome. RESULTS: In multivariate models, women whose diets scored in the highest quintile compared with the lowest quintile on the modified 2005 Healthy Eating Index had 46% lower odds for early AMD. Women in the highest quintile compared with those in the lowest quintile for physical activity (in metabolic energy task hours per week) had 54% lower odds for early AMD. Although smoking was not independently associated with AMD on its own, having a combination of 3 healthy behaviors (healthy diet, physical activity, and not smoking) was associated with 71% lower odds for AMD compared with having high-risk scores (P < .001). CONCLUSION: Modifying lifestyles might reduce risk for early AMD as much as 3-fold, lowering the risk for advanced AMD in a person's lifetime and the social and economic costs of AMD to society.

Healthy diets and the subsequent prevalence of nuclear cataract in women.

OBJECTIVE: To assess the association between healthy diet scores and prevalence of nuclear cataract in women. METHODS: The association between healthy diet scores, which reflect adherence to the US dietary guidelines, and prevalence of nuclear cataract determined 4 to 7 years later was assessed in a sample of Women's Health Initiative Observational Study participants (aged 50-79 years) residing in Iowa, Wisconsin, and Oregon. Scores on the 1995 Healthy Eating Index, which reflect adherence to 1990 guidelines, were assigned from responses to food frequency questionnaires at the Women's Health Initiative baseline (1994-1998). Presence of nuclear cataract was determined from slitlamp photographs and self-reports of cataract extractions were assessed from May 1, 2001, to January 31, 2004, in 1808 women participating in the Carotenoids in Age-Related Eye Disease Study. RESULTS: Having a high 1995 Healthy Eating Index score was the strongest modifiable predictor of low prevalence of nuclear cataract among numerous risk factors investigated in this sample. The multivariate-adjusted odds ratio for high vs low quintile for diet score was 0.63 (95% confidence interval, 0.43-0.91). Higher prevalence of nuclear cataract was also associated with other modifiable factors (smoking and marked obesity) and nonmodifiable factors (having brown eyes, myopia, and high pulse pressure). Vitamin supplement use was not related to cataract. CONCLUSION: These data add to the body of evidence suggesting that eating foods rich in a variety of vitamins and minerals may contribute to postponing the occurrence of the most common type of cataract in the United States.

Women's Health Initiative diet intervention did not increase macular pigment optical density in an ancillary study of a subsample of the Women's Health Initiative.

In this study, we examined the impact of long-term (>8 y), low-fat, high-fruit and -vegetable diets on levels of lutein and zeaxanthin in the macula of the retina, as indicated by the OD of macular pigment. Macular pigment OD, measured by heterochromatic flicker photometry, was compared in women aged 60-87 y, who, 7-18 mo earlier (median 12 mo), had been in the dietary modification intervention (n = 158) or comparison (n = 236) groups of the Women's Health Initiative (WHI) at the Madison, WI site for a mean of 8.5 y. Women in the intervention group ate more fruits and vegetables (mean +/- SEM) (6.1 +/- 0.2 vs. 4.6 +/- 0.2 servings/d; P < 0.0001) and had higher intakes of lutein and zeaxanthin from foods and supplements (2.7 +/- 0.2 vs. 2.1 +/- 0.1 mg/d; P = 0.0003) than the comparison group. However, macular pigment density did not differ between the intervention (0.36 +/- 0.02 OD units) and comparison (0.35 +/- 0.01 OD units) groups. It tended to be higher (11%; P = 0.11) in women consuming lutein and zeaxanthin in the highest compared with the lowest quintile (median 6.4 vs. 1.1 mg/d). The increase in fruit and vegetable intake among dietary modification participants of this WHI subsample was not of sufficient magnitude to alter the mean density of retinal carotenoids, given other existing dietary conditions in this sample.

Associations between age-related nuclear cataract and lutein and zeaxanthin in the diet and serum in the Carotenoids in the Age-Related Eye Disease Study, an Ancillary Study of the Women's Health Initiative.

OBJECTIVE: To evaluate associations between nuclear cataract (determined from slitlamp photographs between May 2001 and January 2004) and lutein and zeaxanthin in the diet and serum in patients between 1994 and 1998 and macula between 2001 and 2004. DESIGN: A total of 1802 women aged 50 to 79 years in Iowa, Wisconsin, and Oregon with intakes of lutein and zeaxanthin above the 78th (high) and below the 28th (low) percentiles in the Women's Health Initiative Observational Study (1994-1998) were recruited 4 to 7 years later (2001-2004) into the Carotenoids in Age-Related Eye Disease Study. RESULTS: Women in the group with high dietary levels of lutein and zeaxanthin had a 23% lower prevalence of nuclear cataract (age-adjusted odds ratio, 0.77; 95% confidence interval, 0.62-0.96) compared with those with low levels. Multivariable adjustment slightly attenuated the association (odds ratio, 0.81; 95% confidence interval, 0.65-1.01). Women in the highest quintile category of diet or serum levels of lutein and zeaxanthin as compared with those in the lowest quintile category were 32% less likely to have nuclear cataract (multivariable-adjusted odds ratio, 0.68; 95% confidence interval, 0.48-0.97; P for trend = .04; and multivariable-adjusted odds ratio, 0.68; 95% confidence interval, 0.47-0.98; P for trend = .01, respectively). Cross-sectional associations with macular pigment density were inverse but not statistically significant. CONCLUSIONS: Diets rich in lutein and zeaxanthin are moderately associated with decreased prevalence of nuclear cataract in older women. However, other protective aspects of such diets may in part explain these relationships.