A non-destructive resonant acoustic testing and defect classification of additively manufactured lattice structures

Additive manufacturing enables the fabrication of lattice structures which are of particular interest to fabricate medical implants and lightweight aerospace parts. Product integrity is critical in these applications. This requests very challenging quality control for such complex geometries, particularly on detecting internal defects. It is important not only to detect whether there are missing struts for a product with a large size of lattices, but also to identify the number of missing struts for safety-critical applications. Resonant ultrasound spectroscopy is a promising method for fast and cost-effective non-destructive testing of complex geometries but data analytics methods are needed to systematically analyze resonant ultrasound signals for defect identification and classification. This study utilizes resonant acoustic method to obtain resonant frequency spectrum of test lattice structures. In addition, regularized linear discriminant analysis, combined with adaptive sampling and normalization, is developed to classify the number of missing struts. The result shows 80.95% testing accuracy on validation study, which suggests that the resonant acoustic method combined with machine learning is a powerful tool to inspect lattices.

Evaluation of Hyperparameter Optimization in Machine and Deep Learning Methods for Decoding Imagined Speech EEG

Classification of electroencephalography (EEG) signals corresponding to imagined speech production is important for the development of a direct-speech brain–computer interface (DS-BCI). Deep learning (DL) has been utilized with great success across several domains. However, it remains an open question whether DL methods provide significant advances over traditional machine learning (ML) approaches for classification of imagined speech. Furthermore, hyperparameter (HP) optimization has been neglected in DL-EEG studies, resulting in the significance of its effects remaining uncertain. In this study, we aim to improve classification of imagined speech EEG by employing DL methods while also statistically evaluating the impact of HP optimization on classifier performance. We trained three distinct convolutional neural networks (CNN) on imagined speech EEG using a nested cross-validation approach to HP optimization. Each of the CNNs evaluated was designed specifically for EEG decoding. An imagined speech EEG dataset consisting of both words and vowels facilitated training on both sets independently. CNN results were compared with three benchmark ML methods: Support Vector Machine, Random Forest and regularized Linear Discriminant Analysis. Intra- and inter-subject methods of HP optimization were tested and the effects of HPs statistically analyzed. Accuracies obtained by the CNNs were significantly greater than the benchmark methods when trained on both datasets (words: 24.97%, p < 1 × 10–7, chance: 16.67%; vowels: 30.00%, p < 1 × 10–7, chance: 20%). The effects of varying HP values, and interactions between HPs and the CNNs were both statistically significant. The results of HP optimization demonstrate how critical it is for training CNNs to decode imagined speech.

Prediction of Protein Tertiary Structure via Regularized Template Classification Techniques

We discuss the use of the regularized linear discriminant analysis (LDA) as a model reduction technique combined with particle swarm optimization (PSO) in protein tertiary structure prediction, followed by structure refinement based on singular value decomposition (SVD) and PSO. The algorithm presented in this paper corresponds to the category of template-based modeling. The algorithm performs a preselection of protein templates before constructing a lower dimensional subspace via a regularized LDA. The protein coordinates in the reduced spaced are sampled using a highly explorative optimization algorithm, regressive–regressive PSO (RR-PSO). The obtained structure is then projected onto a reduced space via singular value decomposition and further optimized via RR-PSO to carry out a structure refinement. The final structures are similar to those predicted by best structure prediction tools, such as Rossetta and Zhang servers. The main advantage of our methodology is that alleviates the ill-posed character of protein structure prediction problems related to high dimensional optimization. It is also capable of sampling a wide range of conformational space due to the application of a regularized linear discriminant analysis, which allows us to expand the differences over a reduced basis set.

Functional brain network modeling in sub-acute stroke patients and healthy controls during rest and continuous attentive tracking

A cerebral stroke is characterized by compromised brain function due to an interruption in cerebrovascular blood supply. Although stroke incurs focal damage determined by the vascular territory affected, clinical symptoms commonly involve multiple functions and cognitive faculties that are insufficiently explained by the focal damage alone. Functional connectivity (FC) refers to the synchronous activity between spatially remote brain regions organized in a network of interconnected brain regions. Functional magnetic resonance imaging (fMRI) has advanced this system-level understanding of brain function, elucidating the complexity of stroke outcomes, as well as providing information useful for prognostic and rehabilitation purposes.We tested for differences in brain network connectivity between a group of patients with minor ischemic strokes in sub-acute phase (n=44) and matched controls (n=100). As neural network configuration is dependent on cognitive effort, we obtained fMRI data during rest and two load levels of a multiple object tacking (MOT) task. Network nodes and time-series were estimated using independent component analysis (ICA) and dual regression, with network edges defined as the partial temporal correlations between node pairs. The full set of edgewise FC went into a cross-validated regularized linear discriminant analysis (rLDA) to classify groups and cognitive load.MOT task performance and cognitive tests revealed no significant group differences. While multivariate machine learning revealed high sensitivity to experimental condition, with classification accuracies between rest and attentive tracking approaching 100%, group classification was at chance level, with negligible differences between conditions. Repeated measures ANOVA showed significantly stronger synchronization between a temporal node and a sensorimotor node in patients across conditions. Overall, the results revealed high sensitivity of FC indices to task conditions, and suggest relatively small brain network-level disturbances after clinically mild strokes.

Developing a Genomics Model to Predict Failure of Isocitrate Dehydrogenase (IDH) Inhibitors for Treatment of Patients with IDH1- or IDH2-Mutated Acute Myeloid Leukemia

Background: Small molecule drugs that target specific pathogenic gene mutations offer the possibility of treating cancers while maintaining a wide therapeutic margin. Potential mechanisms of resistance to such targeted therapies include alternate or additional mutations in the target gene, activating mutations in genes downstream in the same pathway, or mutations in genes in an alternate prosurvival pathway. Co-occurring mutations that contribute to resistance may be present at the start of therapy and lead to immediate treatment failure. Therapy planning might be optimized if one could predict failure to a specific treatment even when the target was present. Ivosidenib (IVO, Agios Pharmaceuticals, Inc.) and enasidenib (ENA, Celgene Corp.) are small molecule inhibitors of IDH1 and IDH2 gain-of-function mutant enzymes, respectively, developed for treatment of acute myeloid leukemia (AML) with such mutations. In the Phase 1 monotherapy trials of IVO and ENA for treatment of relapsed or refractory (R/R) AML, the majority of patients appeared to be resistant to treatment; 67% and 77% of the patients with the target IDH mutation, respectively, failed to achieve complete remission with full (CR) or partial (CRh) hematological recovery. IDH1- and IDH2-mutated AMLs have multiple co-occurring mutations which may impact this resistance. The purpose of this study was to determine if there is a co-occurring mutation signature at treatment baseline that can identify patients with relapsed or refractory IDH1- or IDH2-mutated AML who would be resistant to treatment with mutant IDH inhibitors. We hypothesized that since the molecular pathogenesis was the same in IDH1- and IDH2-mutated AML (overexpression of the oncometabolite 2-hydroxyglutarate ), the co-occurring mutations that contributed to resistance would also be the same. Methods: Adequate clinical and genomics data were available for 147 patients with IDH1-mutated R/R AML treated with IVO (Study AG120-C-001, NCT02074839) and 87 patients with IDH2-mutated R/R AML treated with ENA (Study AG221-C-001, NCT01915498). Only patients with the treatment-targeted IDH mutations as determined by the companion diagnostic were included in the analysis cohorts. Patients who failed to achieve an FDA-adjudicated CR or CRh within the first 6 cycles of therapy were considered nonresponders; patients without a response assessment after start of therapy were excluded from analysis. Genomics data were generated using next-generation sequencing platforms. Results: To address the hypothesis, a mutational signature or model was first generated with the data from the ENA-treated patients, and subsequently tested on the data from the IVO-treated patients. In the first step, data were fit to the model using a proprietary algorithm that implemented a Monte-Carlo simulation expansion of Fisher's regularized linear discriminant analysis. The candidate model with the minimal number of genes that can differentiate more than half of the difference between responder and nonresponder groups includes 19 genes. However, more accuracy was attained by developing a 46- or 96-gene model. By Ingenuity Pathway Analysis, nonresponders differed from responders by mutations in only a few functional groups of genes (leukemia cell proliferation, leukemia cell differentiation and leukemia cell movement). In the assessment of diagnostic accuracy, the 96-gene model had 96% sensitivity, 69% specificity and 90% accuracy for identifying patients with IDH2-mutated AML who would not respond to treatment with ENA (Table). When tested in the cohort of patients with IDH1-mutated AML treated with IVO, the model had 97% sensitivity but only 10% specificity and 65% accuracy. Conclusions: We conclude that genomics data may be useful to identify patients with co-occurring mutations that predict resistance to targeted therapies, but a predictive genomics signature may be specific to the treatment target gene rather than being generalizable across target genes that share a common mechanism of pathogenesis. Table Table. Disclosures No relevant conflicts of interest to declare.