Robustness of Deep Networks for Mammography: Replication Across Public Datasets.

Deep neural networks have demonstrated promising performance in screening mammography with recent studies reporting performance at or above the level of trained radiologists on internal datasets. However, it remains unclear whether the performance of these trained models is robust and replicates across external datasets. In this study, we evaluate four state-of-the-art publicly available models using four publicly available mammography datasets (CBIS-DDSM, INbreast, CMMD, OMI-DB). Where test data was available, published results were replicated. The best-performing model, which achieved an area under the ROC curve (AUC) of 0.88 on internal data from NYU, achieved here an AUC of 0.9 on the external CMMD dataset (N = 826 exams). On the larger OMI-DB dataset (N = 11,440 exams), it achieved an AUC of 0.84 but did not match the performance of individual radiologists (at a specificity of 0.92, the sensitivity was 0.97 for the radiologist and 0.53 for the network for a 1-year follow-up). The network showed higher performance for in situ cancers, as opposed to invasive cancers. Among invasive cancers, it was relatively weaker at identifying asymmetries and was relatively stronger at identifying masses. The three other trained models that we evaluated all performed poorly on external datasets. Independent validation of trained models is an essential step to ensure safe and reliable use. Future progress in AI for mammography may depend on a concerted effort to make larger datasets publicly available that span multiple clinical sites.

Does the Combination of Phone, Email and Text-Based Reminders Improve No-show Rates for Patients in Breast Imaging?

The issue of no-shows in radiology is complicated and challenging. Mammography and ultrasound have the highest rate of no-shows among radiologic exams. Screening mammography is one of the most cost-effective ways to reduce breast cancer related deaths. However, the benefit of screening is heavily dependent on patient compliance to routine exams. Enhancing patients' commitments to their scheduled appointments, thereby improving early detection and decreasing breast cancer related mortality. Retrospective analysis of no-show visits scheduled from August 2017 to December 2017 (before the implementation of combined phone, email and text-based reminders) and from August 2019 to December 2019 (after the implementation of reminder and follow-up phone calls after missed appointments by the coordinator) in an urban academic breast imaging center was conducted. There were 368 no-show patients in 2017 and 238 no-show patients in 2019. Percentage of no-shows, and delay time to the rescheduled missed appointment were calculated. Subgroup analysis of the type of studies that were missed and those who did not reschedule the missed appointment was conducted. Mann Whitney U test was used to analyze differences between group means. No-show visits decreased by 50% in 2019 when compared to 2017. The average wait time between the missed appointment and the rescheduled appointment decreased significantly from 30.7 weeks in 2017 to 12.1 weeks in 2019 (P = 0.047). The percentage of no-show visits was highest among the unemployed, patients scheduled for screening mammograms and patients with a high average of no-show visits. No-show visits adversely impact patient outcome and contribute to increased cost of healthcare. Through a deeper understanding of the factors contributing to no-shows, we can strive to make appropriate interventions to alleviate the consequences of no-shows.

Validation of a novel functional test for assessing metamorphopsia using epiretinal membranes as a model.

Current tests for assessing metamorphopsia do not account for confounders such as perceptual filling-in and spatial redundancy, which affect its sensitivity and repeatability. This proof-of-concept study aimed to assess the performance of a novel laboratory-based psychophysical test (Line Sag Test, LST) which addresses these issues for quantification of metamorphopsia in idiopathic epiretinal membranes. The LST quantifies perpendicular metamorphopsia at three eccentricities (3°, 6°, and 9°) along eight meridians (45° steps). Metamorphopsia was assessed using the LST and Amsler grid and the hit rates of both tests for detecting metamorphopsia were compared. Normal metamorphopsia scores using the LST did not differ significantly from 0 and fell within one step-size (p = 0.500). The LST detected significantly more cases of metamorphopsia than the Amsler grid (14/21 versus 3/21) (p = 0.003). Similarly, significantly more cases of visual distortions in asymptomatic iERMs were detected using the LST than the Amsler grid (11/18 versus 0/18) (p = 0.008). The LST has a higher hit rate compared to the Amsler grid (67% versus 14%). This work demonstrates a psychophysically-robust functional test addressing perceptual confounders is more sensitive for quantifying and localising metamorphopsia in macular disease, particularly in asymptomatic disease.

The Impact of Visual Field Clusters on Performance-based Measures and Vision-Related Quality of Life in Patients With Glaucoma.

PURPOSE: To investigate how visual field (VF) clusters affect performance-based measures of the ability to perform activities of daily living and subjective measures of vision-related quality of life (QoL) in patients with glaucoma. DESIGN: Prospective, cross-sectional study. METHODS: setting: Institutional - Wills Eye Hospital. STUDY POPULATION: 322 eyes of 161 patients with moderate-stage glaucoma. OBSERVATION: VF tests were conducted using the Humphrey 24-2 Swedish Interactive Thresholding Algorithm standard perimeter. The VFs of each patient were divided into 5 clusters: nasal, temporal, central, paracentral, and peripheral. The score for each cluster was the averaged total deviation scores of all tested points within the cluster. Each cluster score was correlated with performance-based measures of visual function and subjective assessment of vision-related QoL. MAIN OUTCOME MEASURES: The Compressed Assessment of Ability Related to Vision, the National Eye Institute Visual Functioning Questionnaire 25 (NEI VFQ-25), and the Modified Glaucoma Symptom Scale. RESULTS: The central VF cluster in the better eye was positively correlated with all Compressed Assessment of Ability Related to Vision (performance-based measure) subscales. The strongest correlation for the better eye was between the central VF cluster and total Compressed Assessment of Ability Related to Vision score (0.39, P < .001). The inferior VF hemisphere in both eyes was positively correlated with most Compressed Assessment of Ability Related to Vision subscales. Central VF clusters in the better eye were positively correlated with a majority of the NEI VFQ-25 subscales. There were no significant correlations between VF clusters and Modified Glaucoma Symptom Scale subscales. CONCLUSIONS: Scores of central VF defects in the better eye and inferior hemisphere defects in both eyes were positively correlated with performance-based measures of the ability to perform activities of daily living. Glaucoma patients with central defects in the better eye were more likely to have reduced scores on assessments of vision-related QoL.

Chongmague reveals an essential role for laminin-mediated boundary formation in chordate convergence and extension movements.

Although cell intercalation driven by non-canonical Wnt/planar cell polarity (PCP) pathway-dependent mediolateral cell polarity is important for notochord morphogenesis, it is likely that multiple mechanisms shape the notochord as it converges and extends. Here we show that the recessive short-tailed Ciona savignyi mutation chongmague (chm) has a novel defect in the formation of a morphological boundary around the developing notochord. chm notochord cells initiate intercalation normally, but then fail to maintain their polarized cell morphology and migrate inappropriately to become dispersed in the larval tail. This is unlike aimless (aim), a mutation in the PCP pathway component Prickle, which has a severe defect in early mediolateral intercalation but forms a robust notochord boundary. Positional cloning identifies chm as a mutation in the C. savignyi ortholog of the vertebrate alpha 3/4/5 family of laminins. Cs-lamalpha3/4/5 is highly expressed in the developing notochord, and Cs-lamalpha3/4/5 protein is specifically localized to the outer border of the notochord. Notochord convergence and extension, reduced but not absent in both chm and aim, are essentially abolished in the aim/aim; chm/chm double mutant, indicating that laminin-mediated boundary formation and PCP-dependent mediolateral intercalation are each able to drive a remarkable degree of tail morphogenesis in the absence of the other. These mechanisms therefore initially act in parallel, but we also find that PCP signaling has an important later role in maintaining the perinotochordal/intranotochordal polarity of Cs-lamalpha3/4/5 localization.