Multi-Objective Performance Optimization of Machine Learning Models in Healthcare.

Multi-objective optimization holds particular significance for medical applications, wherein enhancing sensitivity is crucial to avoid costly missed diagnoses, and maintaining high specificity is imperative to prevent unnecessary procedures. In particular, when optimizing machine learning architectures for clinical diagnostics, it becomes essential to balance target quality measures such as accuracy, sensitivity, and specificity. Therefore, we developed MOOF, a multi-objective optimization framework that employs NSGA-II and TOPSIS to simultaneously optimize the model parameters of three selected ML algorithms: random forest, support vector machine, and multilayer perceptron. Finally, we evaluated the performance of the optimized MOOF models compared to gold standard methods such as multi-score grid search and single objective optimizations. Our results show that MOOF generally outperforms other approaches by inherently providing optimal solutions, representing the trade-offs between the target objectives. In conclusion, the study supports the importance of multi-objective optimization in medical informatics, with MOOF as a powerful tool for precise ML models, potentially improving patient care and clinical decision support systems.

Using Explainable Artificial Intelligence Models (ML) to Predict Suspected Diagnoses as Clinical Decision Support.

The complexity of emergency cases and the number of emergency patients have increased dramatically. Due to a reduced or even missing specialist medical staff in the emergency departments (EDs), medical knowledge is often used without professional supervision for the diagnosis. The result is a failure in diagnosis and treatment, even death in the worst case. Secondary: high expenditure of time and high costs. Using accurate patient data from the German national registry of the medical emergency departments (AKTIN-registry, Home - Notaufnahmeregister (aktin.org)), the most 20 frequent diagnoses were selected for creating explainable artificial intelligence (XAI) models as part of the ENSURE project (ENSURE (umg.eu)). 137.152 samples and 51 features (vital signs and symptoms) were analyzed. The XAI models achieved a mean area under the curve (AUC) one-vs-rest of 0.98 for logistic regression (LR) and 0.99 for the random forest (RF), and predictive accuracies of 0.927 (LR) and 0.99 (RF). Based on its grade of explainability and performance, the best model will be incorporated into a portable CDSS to improve diagnoses and outcomes of ED treatment and reduce cost. The CDSS will be tested in a clinical pilot study at EDs of selected hospitals in Germany.

Development of a Clinical Decision Support System for Smart Algorithms in Emergency Medicine.

The development of clinical decision support systems (CDSS) is complex and requires user-centered planning of assistive interventions. Especially in the setting of emergency care requiring time-critical decisions and interventions, it is important to adapt a CDSS to the needs of the user in terms of acceptance, usability and utility. In the so-called ENSURE project, a user-centered approach was applied to develop the CDSS intervention. In the context of this paper, we present a path to the first mockup development for a CDSS interface by addressing Campbell's Five Rights within the CDSS workflow.

In Vivo Measurements of Cortical Thickness and Porosity at the Proximal Third of the Tibia Using Guided Waves: Comparison with Site-Matched Peripheral Quantitative Computed Tomography and Distal High-Resolution Peripheral Quantitative Computed Tomography.

The aim of this study was to estimate cortical porosity (Ct.Po) and cortical thickness (Ct.Th) using 500-kHz bi-directional axial transmission (AT). Ct.ThAT and Ct.PoAT were obtained at the tibia in 15 patients from a 2-D transverse isotropic free plate model fitted to measured guided wave dispersion curves. The velocities of the first arriving signal (υFAS) and A0 mode (υA0) were also determined. Site-matched peripheral quantitative computed tomography (pQCT) provided volumetric cortical bone mineral density (Ct.vBMDpQCT) and Ct.ThpQCT. Good agreement was found between Ct.ThAT and Ct.ThpQCT (R2 = 0.62, root mean square error [RMSE] = 0.39 mm). Ct.vBMDpQCT correlated with Ct.PoAT (R2 = 0.57), υFAS (R2 = 0.43) and υA0 (R2 = 0.28). Furthermore, a significant correlation was found between AT and distal high-resolution pQCT. The measurement ofcortical parameters at the tibia using guided waves might improve the prediction of bone fractures in a cost-effective and radiation-free manner.

Insights into coronary collateral formation from a novel porcine semiacute infarction model.

BACKGROUND: For patients with severe ischemic heart disease, complete revascularization by a percutaneous coronary intervention or coronary artery bypass grafting is often not achieved and may still cause residual angina. In case of progressive coronary artery occlusions, therapeutic arteriogenesis constitutes a promising strategy for increasing blood supply to the ischemic myocardium. Whether the formation of collaterals in the hypofused myocardium is angiogenetic in nature or based on preformed coronary artery anastomoses remains debatable. The objectives of this research were (i) the development of an appropriate research methodology to study a humanoid animal semiacute infarction model with low mortality and (ii) to answer the question of whether collateral revascularization follows a pre-existing 'blueprint'. MATERIALS AND METHODS: A porcine model was chosen in which a step-wise vessel occlusion was performed by implantation of a copper stent into the distal left anterior descending artery. Vessel occlusion and collateral development were confirmed in vivo every 14 days up to day 56 by repeated coronary angiography and myocardial perfusion measurement using cardiac MRI. After the completion of the in-vivo imaging studies, animals were euthanized and collateral growth was evaluated using microcomputer tomography. RESULTS: Our porcine model of semiacute noninvasive coronary artery occlusion confirmed the existence of preformed coronary anastomoses and the proliferation of functional vessels in hypoperfused myocardium. Repetitive intra-animal MRIs showed the functional impact of these growing collaterals. CONCLUSION: The confirmation of preformed coronary anastomoses during the process of collateralization (natural bypasses) offers a preclinical avenue to carry out arteriogenetic pharmaceutical research in patients with ischemic heart disease.

Porcine arteriogenesis based on vasa vasorum in a novel semi-acute occlusion model using high-resolution imaging.

Bridging collaterals (BC) develop in several chronic total artery occlusion diseases, and can prevent extensive myocardial necrosis. Yet, their origin, growth process, and histo-morphology are still unclear. Since vasa vasorum (VV) may take part in collateralization, we hypothesized that VV are the basis for BCs. To comprehensively investigate this arteriogenesis process, we used high-resolution imaging, including corrosion casts, post-mortem angiography with stereoscopy, micro-CT, and immunohistology, in combination with a novel semi-acute vessel occlusion model. This porcine model was produced by implanting a copper stent minimally invasively into the left anterior descending coronary artery. To define the kinetics of arteriogenesis, pigs (n = 11) were assigned to one of the five euthanasia timepoints: day 0.5 (D0.5, n = 2), D3 (n = 2), D5 (n = 1), D7 (n = 3), or D12 (n = 3) after stent implantation. We found that (1) BCs originate from longitudinally running type 1 VV, mainly VV interna, partially also from VV externa; (2) the growth of VV to BC is rapid, occurring within 7 days; and (3) porcine BCs are likely functionally relevant, considering an observed 102% increase in the number of smooth muscle cell layers in their vascular wall. High-resolution imaging in a minimally invasive non-acute vessel occlusion model is an innovative technique that allowed us to provide direct evidence that porcine BCs develop from the VV. These data may be crucial for further studies on the treatment of angina pectoris and thromboangiitis obliterans through therapeutic stimulation of BC development.

Knockout of Density-Enhanced Phosphatase-1 impairs cerebrovascular reserve capacity in an arteriogenesis model in mice.

Collateral growth, arteriogenesis, represents a proliferative mechanism involving endothelial cells, smooth muscle cells, and monocytes/macrophages. Here we investigated the role of Density-Enhanced Phosphatase-1 (DEP-1) in arteriogenesis in vivo, a protein-tyrosine-phosphatase that has controversially been discussed with regard to vascular cell biology. Wild-type C57BL/6 mice subjected to permanent left common carotid artery occlusion (CCAO) developed a significant diameter increase in distinct arteries of the circle of Willis, especially in the anterior cerebral artery. Analyzing the impact of loss of DEP-1 function, induction of collateralization was quantified after CCAO and hindlimb femoral artery ligation comparing wild-type and DEP-1(-/-) mice. Both cerebral collateralization assessed by latex perfusion and peripheral vessel growth in the femoral artery determined by microsphere perfusion and micro-CT analysis were not altered in DEP-1(-/-) compared to wild-type mice. Cerebrovascular reserve capacity, however, was significantly impaired in DEP-1(-/-) mice. Cerebrovascular transcriptional analysis of proarteriogenic growth factors and receptors showed specifically reduced transcripts of PDGF-B. SiRNA knockdown of DEP-1 in endothelial cells in vitro also resulted in significant PDGF-B downregulation, providing further evidence for DEP-1 in PDGF-B gene regulation. In summary, our data support the notion of DEP-1 as positive functional regulator in vascular cerebral arteriogenesis, involving differential PDGF-B gene expression.

Pulsatile shear and Gja5 modulate arterial identity and remodeling events during flow-driven arteriogenesis.

In the developing chicken embryo yolk sac vasculature, the expression of arterial identity genes requires arterial hemodynamic conditions. We hypothesize that arterial flow must provide a unique signal that is relevant for supporting arterial identity gene expression and is absent in veins. We analyzed factors related to flow, pressure and oxygenation in the chicken embryo vitelline vasculature in vivo. The best discrimination between arteries and veins was obtained by calculating the maximal pulsatile increase in shear rate relative to the time-averaged shear rate in the same vessel: the relative pulse slope index (RPSI). RPSI was significantly higher in arteries than veins. Arterial endothelial cells exposed to pulsatile shear in vitro augmented arterial marker expression as compared with exposure to constant shear. The expression of Gja5 correlated with arterial flow patterns: the redistribution of arterial flow provoked by vitelline artery ligation resulted in flow-driven collateral arterial network formation and was associated with increased expression of Gja5. In situ hybridization in normal and ligation embryos confirmed that Gja5 expression is confined to arteries and regulated by flow. In mice, Gja5 (connexin 40) was also expressed in arteries. In the adult, increased flow drives arteriogenesis and the formation of collateral arterial networks in peripheral occlusive diseases. Genetic ablation of Gja5 function in mice resulted in reduced arteriogenesis in two occlusion models. We conclude that pulsatile shear patterns may be central for supporting arterial identity, and that arterial Gja5 expression plays a functional role in flow-driven arteriogenesis.