David Lehmann
Postdoctoral Scholar, Cardiovascular Medicine
Bio
Dr. David Hermann Lehmann is a Postdoctoral Researcher at Stanford University School of Medicine, where his research focuses on artificial intelligence for cardiovascular medicine. His work lies at the intersection of computer vision, machine learning, and clinical cardiology, with a focus on developing AI methods for automated cardiovascular diagnosis by integrating multimodal medical data, including cardiac magnetic resonance imaging (MRI), electrocardiograms (ECGs), and electronic health records (EHRs). He also develops medical AI agents that leverage these diverse data sources to support clinical reasoning and advance intelligent, data-driven healthcare.
His research aims to translate advances in AI into clinically meaningful applications, improving the accuracy, scalability, and accessibility of cardiovascular diagnostics. His work spans multimodal medical data analysis, deep learning for cardiac imaging, and intelligent systems for clinical decision support.
He earned his doctoral degree (Dr. rer. nat.) in computer science through research at the Engineering Mathematics Computing Lab and the Cardiology Department of Heidelberg University Hospital. His doctoral work focused on AI-based diagnosis of cardiac diseases using MRI and smartwatch ECGs. He also holds a master’s degree in computer science from Technische Universität Darmstadt and a bachelor’s degree in physics from the same university.
Professional Education
-
Doctor of Science, University of Heidelberg, Computer Science (2024)
-
Master of Science, Technische Universitat Darmstadt, Computer Science (2019)
-
Bachelor of Science, Technische Universitat Darmstadt, Physics (2016)
All Publications
-
Prediction of diagnosis and diastolic filling pressure by AI-enhanced cardiac MRI: a modelling study of hospital data.
The Lancet. Digital health
2024; 6 (6): e407-e417
Abstract
With increasing numbers of patients and novel drugs for distinct causes of systolic and diastolic heart failure, automated assessment of cardiac function is important. We aimed to provide a non-invasive method to predict diagnosis of patients undergoing cardiac MRI (cMRI) and to obtain left ventricular end-diastolic pressure (LVEDP).For this modelling study, patients who had undergone cardiac catheterisation at University Hospital Heidelberg (Heidelberg, Germany) between July 15, 2004 and March 16, 2023, were identified, as were individual left ventricular pressure measurements. We used existing patient data from routine cardiac diagnostics. From this initial group, we extracted patients who had been diagnosed with ischaemic cardiomyopathy, dilated cardiomyopathy, hypertrophic cardiomyopathy, or amyloidosis, as well as control individuals with no structural phenotype. Data were pseudonymised and only processed within the university hospital's AI infrastructure. We used the data to build different models to predict either demographic (ie, AI-age and AI-sex), diagnostic (ie, AI-coronary artery disease and AI-cardiomyopathy [AI-CMP]), or functional parameters (ie, AI-LVEDP). We randomly divided our datasets via computer into training, validation, and test datasets. AI-CMP was not compared with other models, but was validated in a prospective setting. Benchmarking was also done.66 936 patients who had undergone cardiac catheterisation at University Hospital Heidelberg were identified, with more than 183 772 individual left ventricular pressure measurements. We extracted 4390 patients from this initial group, of whom 1131 (25·8%) had been diagnosed with ischaemic cardiomyopathy, 1064 (24·2%) had been diagnosed with dilated cardiomyopathy, 816 (18·6%) had been diagnosed with hypertrophic cardiomyopathy, 202 (4·6%) had been diagnosed with amyloidosis, and 1177 (26·7%) were control individuals with no structural phenotype. The core cohort only included patients with cardiac catherisation and cMRI within 30 days, and emergency cases were excluded. AI-sex was able to predict patient sex with areas under the receiver operating characteristic curves (AUCs) of 0·78 (95% CI 0·77-0·78) and AI-age was able to predict patient age with a mean absolute error of 7·86 years (7·77-7·95), with a Pearson correlation of 0·57 (95% CI 0·56-0·57). The AUCs for the classification tasks ranged between 0·82 (95% CI 0·79-0·84) for ischaemic cardiomyopathy and 0·92 (0·91-0·94) for hypertrophic cardiomyopathy.Our AI models could be easily integrated into clinical practice and provide added value to the information content of cMRI, allowing for disease classification and prediction of diastolic function.Informatics for Life initiative of the Klaus-Tschira Foundation, German Center for Cardiovascular Research, eCardiology section of the German Cardiac Society, and AI Health Innovation Cluster Heidelberg.
View details for DOI 10.1016/S2589-7500(24)00063-3
View details for PubMedID 38789141
-
Improving sudden cardiac death risk stratification in hypertrophic cardiomyopathy using established clinical variables and genetic information
CLINICAL RESEARCH IN CARDIOLOGY
2024; 113 (5): 728-736
Abstract
The cardiac societies of Europe and the United States have established different risk models for preventing sudden cardiac death (SCD) in hypertrophic cardiomyopathy (HCM). The aim of this study is to validate current SCD risk prediction methods in a German HCM cohort and to improve them by the addition of genotype information.HCM patients without prior SCD or equivalent arrhythmic events ≥ 18 years of age were enrolled in an expert cardiomyopathy center in Germany. The primary endpoint was defined as SCD/-equivalent within 5 years of baseline evaluation. 5-year SCD-risk estimates and recommendations for ICD implantations, as defined by the ESC and AHA/ACC guidelines, were analyzed. Multivariate cox proportional hazards analyses were integrated with genetic findings as additive SCD risk.283 patients were included and followed for in median 5.77 years (2.92; 8.85). A disease-causing variant was found in 138 (49%) patients. 14 (5%) patients reached the SCD endpoint (5-year incidence 4.9%). Kaplan-Meier survival analysis shows significantly lower overall SCD event-free survival for patients with an identified disease-causing variant (p < 0.05). The ESC HCM Risk-SCD model showed an area-under-the-curve (AUC) of 0.74 (95% CI 0.68-0.79; p < 0.0001) with a sensitivity of 0.29 (95% CI 0.08-0.58) and specificity of 0.83 (95% CI 0.78-0.88) for a risk estimate ≥ 6%/5-years. By comparison, the AHA/ACC HCM SCD risk stratification model showed an AUC of 0.70 (95% CI 0.65-0.76; p = 0.003) with a sensitivity of 0.93 (95% CI, 0.66-0.998) and specificity of 0.28 (95% CI 0.23-0.34) at the respective cut-off. The modified SCD Risk Score with genetic information yielded an AUC of 0.76 (95% CI 0.71-0.81; p < 0.0001) with a sensitivity of 0.86 (95% CI 0.57-0.98) and specificity of 0.69 (95% CI 0.63-0.74). The number-needed-to-treat (NNT) to prevent 1 SCD event by prophylactic ICD-implantation is 13 for the ESC model, 28 for AHA/ACC and 9 for the modified Genotype-model.This study confirms the performance of current risk models in clinical decision making. The integration of genetic findings into current SCD risk stratification methods seem feasible and can add in decision making, especially in borderline risk-groups. A subgroup of patients with high SCD risk remains unidentified by current risk scores.
View details for DOI 10.1007/s00392-023-02310-4
View details for Web of Science ID 001079905500001
View details for PubMedID 37792019
View details for PubMedCentralID PMC11026183
-
Multi-omics assessment of dilated cardiomyopathy using non-negative matrix factorization
PLOS ONE
2022; 17 (8): e0272093
Abstract
Dilated cardiomyopathy (DCM), a myocardial disease, is heterogeneous and often results in heart failure and sudden cardiac death. Unavailability of cardiac tissue has hindered the comprehensive exploration of gene regulatory networks and nodal players in DCM. In this study, we carried out integrated analysis of transcriptome and methylome data using non-negative matrix factorization from a cohort of DCM patients to uncover underlying latent factors and covarying features between whole-transcriptome and epigenome omics datasets from tissue biopsies of living patients. DNA methylation data from Infinium HM450 and mRNA Illumina sequencing of n = 33 DCM and n = 24 control probands were filtered, analyzed and used as input for matrix factorization using R NMF package. Mann-Whitney U test showed 4 out of 5 latent factors are significantly different between DCM and control probands (P<0.05). Characterization of top 10% features driving each latent factor showed a significant enrichment of biological processes known to be involved in DCM pathogenesis, including immune response (P = 3.97E-21), nucleic acid binding (P = 1.42E-18), extracellular matrix (P = 9.23E-14) and myofibrillar structure (P = 8.46E-12). Correlation network analysis revealed interaction of important sarcomeric genes like Nebulin, Tropomyosin alpha-3 and ERC-protein 2 with CpG methylation of ATPase Phospholipid Transporting 11A0, Solute Carrier Family 12 Member 7 and Leucine Rich Repeat Containing 14B, all with significant P values associated with correlation coefficients >0.7. Using matrix factorization, multi-omics data derived from human tissue samples can be integrated and novel interactions can be identified. Hypothesis generating nature of such analysis could help to better understand the pathophysiology of complex traits such as DCM.
View details for DOI 10.1371/journal.pone.0272093
View details for Web of Science ID 000892314200025
View details for PubMedID 35980883
View details for PubMedCentralID PMC9387871
-
A novel risk model for predicting potentially life-threatening arrhythmias in non-ischemic dilated cardiomyopathy (DCM-SVA risk).
International journal of cardiology
2021; 339: 75-82
Abstract
Non-ischemic dilated cardiomyopathy (DCM) can be complicated by sustained ventricular arrhythmias (SVA) and sudden cardiac death (SCD). By now, left-ventricular ejection fraction (LV-EF) is the main guideline criterion for primary prophylactic ICD implantation, potentially leading either to overtreatment or failed detection of patients at risk without severely impaired LV-EF. The aim of the European multi-center study DETECTIN-HF was to establish a clinical risk calculator for individualized risk stratification of DCM patients.1393 patients (68% male, mean age 50.7 ± 14.3y) from four European countries were included. The outcome was occurrence of first potentially life-threatening ventricular arrhythmia. The model was developed using Cox proportional hazards, and internally validated using cross validation. The model included seven independent and easily accessible clinical parameters sex, history of non-sustained ventricular tachycardia, history of syncope, family history of cardiomyopathy, QRS duration, LV-EF, and history of atrial fibrillation. The model was also expanded to account for presence of LGE as the eight8h parameter for cases with available cMRI and scar information.During a mean follow-up period of 57.0 months, 193 (13.8%) patients experienced an arrhythmic event. The calibration slope of the developed model was 00.97 (95% CI 0.90-1.03) and the C-index was 0.72 (95% CI 0.71-0.73). Compared to current guidelines, the model was able to protect the same number of patients (5-year risk ≥8.5%) with 15% fewer ICD implantations.This DCM-SVA risk model could improve decision making in primary prevention of SCD in non-ischemic DCM using easily accessible clinical information and will likely reduce overtreatment.
View details for DOI 10.1016/j.ijcard.2021.07.002
View details for PubMedID 34245791
-
microRNA neural networks improve diagnosis of acute coronary syndrome (ACS)
JOURNAL OF MOLECULAR AND CELLULAR CARDIOLOGY
2021; 151: 155-162
Abstract
Cardiac troponins are the preferred biomarkers of acute myocardial infarction. Despite superior sensitivity, serial testing of Troponins to identify patients suffering acute coronary syndromes is still required in many cases to overcome limited specificity. Moreover, unstable angina pectoris relies on reported symptoms in the troponin-negative group. In this study, we investigated genome-wide miRNA levels in a prospective cohort of patients with clinically suspected ACS and determined their diagnostic value by applying an in silico neural network.PAXgene blood and serum samples were drawn and hsTnT was measured in patients at initial presentation to our Chest-Pain Unit. After clinical and diagnostic workup, patients were adjudicated by senior cardiologists in duty to their final diagnosis: STEMI, NSTEMI, unstable angina pectoris and non-ACS patients. ACS patients and a cohort of healthy controls underwent deep transcriptome sequencing. Machine learning was implemented to construct diagnostic miRNA classifiers.We developed a neural network model which incorporates 34 validated ACS miRNAs, showing excellent classification results. By further developing additional machine learning models and selecting the best miRNAs, we achieved an accuracy of 0.96 (95% CI 0.96-0.97), sensitivity of 0.95, specificity of 0.96 and AUC of 0.99. The one-point hsTnT value reached an accuracy of 0.89, sensitivity of 0.82, specificity of 0.96, and AUC of 0.96.Here we show the concept of neural network based biomarkers for ACS. This approach also opens the possibility to include multi-modal data points to further increase precision and perform classification of other ACS differential diagnoses.
View details for DOI 10.1016/j.yjmcc.2020.04.014
View details for Web of Science ID 000617923800011
View details for PubMedID 32305360
-
Epigenetic Regulation of Alternative mRNA Splicing in Dilated Cardiomyopathy
JOURNAL OF CLINICAL MEDICINE
2020; 9 (5)
Abstract
In recent years, the genetic architecture of dilated cardiomyopathy (DCM) has been more thoroughly elucidated. However, there is still insufficient knowledge on the modifiers and regulatory principles that lead to the failure of myocardial function. The current study investigates the association of epigenome-wide DNA methylation and alternative splicing, both of which are important regulatory principles in DCM. We analyzed screening and replication cohorts of cases and controls and identified distinct transcriptomic patterns in the myocardium that differ significantly, and we identified a strong association of intronic DNA methylation and flanking exons usage (p < 2 × 10-16). By combining differential exon usage (DEU) and differential methylation regions (DMR), we found a significant change of regulation in important sarcomeric and other DCM-associated pathways. Interestingly, inverse regulation of Titin antisense non-coding RNA transcript splicing and DNA methylation of a locus reciprocal to TTN substantiate these findings and indicate an additional role for non-protein-coding transcripts. In summary, this study highlights for the first time the close interrelationship between genetic imprinting by DNA methylation and the transport of this epigenetic information towards the dynamic mRNA splicing landscape. This expands our knowledge of the genome-environment interaction in DCM besides simple gene expression regulation.
View details for DOI 10.3390/jcm9051499
View details for Web of Science ID 000540223800256
View details for PubMedID 32429430
View details for PubMedCentralID PMC7291244