Geometric deep learning provides a principled and versatile manner for integration of imaging and non-imaging modalities in the medical domain. Graph Convolutional Networks (GCNs) in particular have been explored on a wide variety of problems such as disease prediction, segmentation, and matrix completion by leveraging large, multimodal datasets. In this paper, we introduce a new spectral domain architecture for deep learning on graphs for disease prediction. The novelty lies in defining geometric 'inception modules' which are capable of capturing intra-and inter-graph structural heterogeneity during convolutions. We design filters with different kernel sizes to build our architecture. We show our disease prediction results on two publicly available datasets. Further, we provide insights on the behaviour of regular GCNs and our proposed model under varying input scenarios on simulated data.

No abstract

In a large mid-European study population including subgroup analysis of mild and moderate keratoconus, accelerated CXL showed similar results to conventional CXL regarding keratometry, corneal topography indices, and CDVA, but further improvement of UDVA. Preoperative Kmax did not affect the postoperative course of corneal topography indices and TCT.

Purpose To compare two calculators for toric intraocular lens (IOL) calculation and to evaluate the prediction of refractive outcome. Methods Sixty-four eyes of forty-five patients underwent cataract surgery followed by implantation of a toric intraocular lens (Zeiss Torbi 709 M) calculated by a standard industry calculator using front keratometry values. Prediction error, median absolute error, and refractive astigmatism error were evaluated for the standard calculator. The predicted postoperative refraction and toric lens power values were evaluated and compared after postoperative recalculation using the Barrett calculator. Results We observed a significant undercorrection in the spherical equivalent (0.19 D) by using a standard calculator (p ≤ 0.05). According to the Baylor nomogram and the refractive influence of posterior corneal astigmatism (PCA), undercorrection of the cylinder was lower for patients with WTR astigmatism, because of the tendency of overcorrection. An advantage of less residual postoperative SE, sphere, and cylinder for the Barrett calculator was observed when retrospectively comparing the calculated predicted postoperative refraction between calculators (p ≤ 0.01). Conclusion Consideration of only corneal front keratometric values for toric lens calculation may lead to postoperative undercorrection of astigmatism. The prediction of postoperative refractive outcome can be improved by using appropriate methods of adjustment in order to take PCA into account.

No abstract

PurposeDifferent modalities of radiation therapy nowadays allow for effective treatment of uveal melanoma combined with the advantage of eye preservation. However, this advantage can secondarily be impaired by radiation‐related side effects. After local recurrence, secondary glaucoma (SG) has been described as second most frequent complication leading to need of enucleation. This study compares the incidence of SG after conventional Ruthenium (Ru)‐106 brachytherapy (BT) versus CyberKnife robotic radiosurgery (RRS) which has been gaining importance lately as an efficient treatment option offering improved patient comfort.MethodsMedical records of all patients diagnosed with uveal melanoma in the Eye Clinic of the Ludwig‐Maximilians‐University Munich between 2007 and 2013 were reviewed. A total of 268 eyes of 268 patients treated with Ru‐106 BT or CyberKnife‐RRS as monotherapy were entered in this retrospective cohort study. Incidence of SG was correlated with treatment modality and baseline tumour characteristics.ResultsFifty‐three patients (19.8%) developed SG. At 5 years, SG was significantly more frequent after RRS (46.7%) than BT (11.1%); however, tumour thickness (maximum apical height) as a marker of tumour progress was more pronounced in the RRS group. Subgroup analysis of 178 patients for tumours amenable to both BT and RRS (thickness ≤6 mm) revealed comparable results at 3 years (RRS: 13.8 versus BT: 11.2%), but a trend towards increased incidence after RRS beyond year three. However, this difference was not significant at 5 years (28.2% versus 11.2%, p = 0.138). Tumour thickness was significantly associated with incidence of SG.ConclusionIn tumours ≤6 mm thickness, RRS and BT seem to offer a comparable safety profile in terms of SG. Beyond year three, SG was tendentially, but not significantly more frequent after RRS. Increasing tumour thickness is associated with risk of SG.

BackgroundA specific Electronic Health Record (EHR) for ophthalmology was introduced in an academic center in Germany. As diagnoses coding corresponding to the International Classification of Diseases Version 10 (ICD-10) is mandatory for billing reasons in Germany, we analyzed whether a change occurred in the diversity and number of diagnoses after the EHR introduction. The number of patients was also analyzed. Proper diagnoses coding is of the utmost importance for further data analysis or billing.MethodsGraphical User Interfaces (GUIs) were created by using Advanced Business Application Programming language in EHR “i.s.h.med.” Development of an EHR was conducted in close collaboration between physicians and software engineers. ICD-10 coding was implemented by using a “hit list” and a search engine for diagnoses. An observational analysis of a 6-month period prior to and after the introduction of an ophthalmological specific EHR was conducted by investigating the diversity and number of diagnoses in various ophthalmological disease categories and the number of patient consultations.ResultsDuring the introduction of a specific ophthalmological EHR, we observed a significant increase in the emergency department cases (323.9 vs. 359.9 cases per week), possibly related to documentation requirements. The number of scheduled outpatients didn’t change significantly (355.12 vs. 360.24 cases per week). The variety of diagnoses also changed: on average, 156.2 different diagnoses were made per week throughout our hospital before the EHR launch, compared to 186.8 different diagnoses per week thereafter (p < 0.05). Additionally, a significantly higher number of diagnoses per case and per week were observed in both emergency and subspecialty outpatient clinics (1.15 vs. 1.22 and 1.10 vs. 1.47, respectively).ConclusionsAn optimized EHR was created for ophthalmological needs and for simplified ICD-10 coding. The implementation of digital patient recording increased the diversity of the diagnoses used per case as well as the number of diagnoses coded per case. A general limitation to date is the suboptimal precision of ICD-10 coding in ophthalmology. Correct coding is of utmost importance for future data analysis.Electronic supplementary materialThe online version of this article (doi:10.1186/s12911-016-0340-1) contains supplementary material, which is available to authorized users.

Background We introduced a video consultation (VC) during the coronavirus (COVID-19) pandemic in an ophthalmology practice with eight doctors to ensure continuous ophthalmological care, infection prophylaxis and to compensate a decreased number of patient presentations. Objective Evaluation of the most common reasons for patient presentations in the VC, the proportion of re-presentations in the practice despite VC, practical challenges associated with the introduction of VC and patient satisfaction. Material and methods Patients with a recent acute visual deterioration and severe eye pain were excluded from the VC. The VC were carried out by a trained specialist in ophthalmology. A questionnaire with eight questions was completed after the VC appointment in order to evaluate the proportion of completed VC and patient satisfaction. Results We included 29 (13 male, Ø 52.6 years, 16 female, Ø 64.7 years) patients in this analysis. The VC could be performed with 68.97% of the participants who rated their overall experience with an average grade of 1.6 (1 very good to 6 insufficient) and all of them indicated that they would recommend the VC. Of presentations in VC 70% were related to the symptoms of the anterior eye segment. In 70% of the cases no re-presentations took place in the unit. Conclusion Our study represents a significant practical application of VC for the management of non-urgent ocular conditions with maximum infection prophylaxis. The introduction of VC was severely limited by technological or user-related issues by the establishment of video connections. Patient satisfaction with VC was high to very high.