Abstract:
Cranio-maxillofacial surgery often alters the aesthetics of the face which can be a heavy burden for patients to decide whether or not to undergo surgery. Today, physicians can predict the post-operative face using surgery planning tools to support the patient’s decision-making. While these planning tools allow a simulation of the post-operative face, the facial texture must usually be captured by another 3D texture scan and subsequently mapped on the simulated face. This approach often results in face predictions that do not appear realistic or lively looking and are therefore ill-suited to guide the patient’s decision-making. Instead, we propose a method using a generative adversarial network to modify a facial image according to a 3D soft-tissue estimation of the post-operative face. To circumvent the lack of available data pairs between pre- and post-operative measurements we propose a semi-supervised training strategy using cycle losses that only requires paired open-source data of images and 3D surfaces of the face’s shape. After training on “in-the-wild” images we show that our model can realistically manipulate local regions of a face in a 2D image based on a modified 3D shape. We then test our model on four clinical examples where we predict the post-operative face according to a 3D soft-tissue prediction of surgery outcome, which was simulated by a surgery planning tool. As a result, we aim to demonstrate the potential of our approach to predict realistic post-operative images of faces without the need of paired clinical data, physical models, or 3D texture scans.
Abstract:
Aims Chronic left atrial enlargement (LAE) increases the risk of atrial fibrillation. Electrocardiogram (ECG) criteria might provide a means to diagnose LAE and identify patients at risk; however, current criteria perform poorly. We seek to characterize the potentially differential effects of atrial dilation vs. hypertrophy on the ECG P-wave. Methods and results We predict effects on the P-wave of (i) left atrial dilation (LAD), i.e. an increase of LA cavity volume without an increase in myocardial volume, (ii) left atrial concentric hypertrophy (LACH), i.e. a thickened myocardial wall, and (iii) a combination of the two. We performed a computational study in a cohort of 72 anatomical variants, derived from four human atrial anatomies. To model LAD, pressure was applied to the LA endocardium increasing cavity volume by up to 100%. For LACH, the LA wall was thickened by up to 3.3 mm. P-waves were derived by simulating atrial excitation propagation and computing the body surface ECG. The sensitivity regarding changes beyond purely anatomical effects was analysed by altering conduction velocity by 25% in 96 additional model variants. Left atrial dilation prolonged P-wave duration (PWd) in two of four subjects; in one subject a shortening, and in the other a variable change were seen. Left atrial concentric hypertrophy, in contrast, consistently increased P-wave terminal force in lead V1 (PTF-V1) in all subjects through an enlarged amplitude while PWd was unaffected. Combined hypertrophy and dilation generally enhanced the effect of hypertrophy on PTF-V1. Conclusion Isolated LAD has moderate effects on the currently used P-wave criteria, explaining the limited utility of PWd and PTF-V1 in detecting LAE in clinical practice. In contrast, PTF-V1 may be a more sensitive indicator of LA myocardial hypertrophy.
Abstract:
P-wave assessment is frequently used in clinical practice to recognize atrial abnormalities. However, the use of P-wave criteria to diagnose specific atrial abnormalities such as left atrial enlargement has shown to be of limited use since these abnormalities can be difficult to distinguish using P-wave criteria to date. Hence, a mechanistic understanding how specific atrial abnormalities affect the P-wave is desirable. In this study, we investigated the effect of left atrial hypertrophy on P-wave morphology using an in silico approach. In a cohort of four realistic patient models, we homogeneously increased left atrial wall thickness in up to seven degrees of left atrial hypertrophy. Excitation conduction was simulated using a monodomain finite element approach. Then, the resulting transmembrane voltage distribution was used to calculate the corresponding extracellular potential distribution on the torso by solving the forward problem of electrocardiography. In our simulation setup, left atrial wall thickening strongly correlated with an increased absolute value of the P-wave terminal force (PTF) in Wilson lead V1 due to an increased negative amplitude while P-wave duration was unaffected. Remarkably, an increased PTF-V1 has often been associated with left atrial enlargement which is defined as a rather increased left atrial volume than a solely thickened left atrium. Hence, the observed contribution of left atrial wall thickness changes to PTF-V1 might explain the poor empirical correlation of left atrial enlargement with PTF-V1.
Abstract:
P-wave morphology correlates with the risk for atrial fibrillation (AF). Left atrial (LA) enlargement could ex- plain both the higher risk for AF and higher P-wave ter- minal force (PTF) in ECG lead V1. However, PTF-V1 has been shown to correlate poorly with LA size. We hypoth- esize that LA hypertrophy, i.e. a thickening of the myocar- dial wall, also contributes to increased PTF-V1 and is part of the reason for the rather low specificity of increased PTF-V1 regarding LA enlargement. To show this, atrial excitation propagation was simulated in a cohort of four anatomically individualized models in- cluding rule-based myocyte orientation and spatial elec- trophysiological heterogeneity using the monodomain ap- proach. The LA wall was thickened symmetrically in steps of 0.66 mm by up to 3.96 mm. Interatrial conduction was possible via discrete connections at the coronary sinus, Bachmann’s bundle and posteriorly. Body surface ECGs were computed using realistic, heterogeneous torso mod- els. During the early P-wave stemming from sources in the RA, no changes were observed. Once the LA got activated, the voltage in V1 tended to lower values for higher degrees of hypertrophy. Thus, the amplitude of the late positive P- wave decreased while the amplitude of the subsequent neg- ative terminal phase increased. PTF-V1 and LA wall thick- ening showed a correlation of 0.95. The P-wave duration was almost unaffected by LA wall thickening (∆ ≤2 ms). Our results show that PTF-V1 is a sensitive marker for LA wall thickening and elucidate why it is superior to P-wave area. The interplay of LA hypertrophy and dilation might cause the poor empirical correlation of LA size and PTF- V1.
Abstract:
Prediction of the post-operative face after craniofacial surgery is an import tool to guide the patient’s decision making to undergo surgery. Hereby, neural networks might provide an inexpensive and fast means to predict the post-operative face without the need for expensive tomography or 3D scans compared to existing methods for surgery prediction. Ideally, these neural networks would be able to predict the post-operative face based on images of the patient and a virtual plan to describe craniofacial surgery. To solve tasks, neural networks have to learn parameterized models directly from data. However, learning of neural networks to predict post-operative faces using supervision would be unfeasible since paired data of pre- and corresponding post-measurements are only sparsely available and often comprise a large time gap of several month between measurements. On the other hand, the invention of CycleGANs has enabled a means to map measurements of one domain to measurements of another domain without requiring paired data for training. This mapping can be derived by enforcing the domain-specific statistics in the predicted measurement and penalizing the reconstruction loss after back and forth translation between two domains. An application of CycleGANs to predict post-operative faces in a clinical setup would require accurate and realistic projections of varying, virtual surgery plans on the pre-operative face. To analyse the suitability of CycleGANs to meet these requirements, a strongly simplified setup was proposed in this work. Hereby, a CycleGAN was developed to predict the outcome after applying four different virtual plans of a modified face in 3D space to a 2D image of a face. For simplification, these virtual plans to describe a modification were represented by a statistical model of a 3D face and were not required to reflect clinical surgery plans. After teaching a CycleGAN to modify an image based on a virtual 3D plan, the resulting images were shown to ten volunteers who were asked to recognize the applied modification in the image and to judge whether the implementation of the modification was realistic or not. In total, the volunteers correctly recognized the applied modifications in 47.33% of images of which 18.0% were judged to be realistic as well. Additionally, the preliminary results of this work suggested the ability of the proposed CycleGAN to interpolate between virtual plans to describe a modified face. As a conclusion, the proposed CylceGAN showed its potential to predict realistically modified faces that satisfied the requirements of the strongly simplified setup in this work. On the other hand, the robustness of the implemented CycleGAN was low i.e. the applied modifications were often not recognizable or not realistic. The results of this work offer an initial attempt to analyze the suitability of CycleGANs to predict post-operative faces.
Abstract:
P-wave assessment can offer a simple and inexpensive means to diagnose left atrial enlargement, which is a predictor for atrial fibrillation. However, the underlying influence of left atrial enlargement on the P-wave is not fully understood. Furthermore, P-wave markers to assess left atrial enlargement show poor sensitivity or specificity - potentially due to other left atrial abnormalities similarly affecting P-wave morphology as left atrial enlargement. In an in silico approach, the left atrium was dilated by mechanical inflation. In another approach, the left atrial myocardium was homogeneously thickened in four patient models. To generate torso meshes containing the modified atria, an existing method to mesh tetrahedral torsos was applied for left atrial wall thickening. For left atrial dilation, a new method was developed to generate tetrahedral torso meshes from organ surfaces. Afterwards, the P-waves were simulated for the two different left atrial anatomical alterations. Left atrial wall thickening resulted in an increased P-terminal force in V1 and an increased P-wave area. For left atrial dilation, the resulting P-waves were not realistic, as the underlying electrophysiological simulation method was not es- tablished yet. Furthermore, left atrial dilation was only applied to a single model. Thus, the results were unreliable. Considering this, the results suggest a prolonged P-wave duration, P-wave notching and an increased P-wave amplitude in aV L to be correlated with left atrial dilation.