Browsing by Author "Fuentes, David"
Now showing 1 - 4 of 4
Results Per Page
Sort Options
Item A molecular dynamics approach towards evaluating osmotic and thermal stress in the extracellular environment(Taylor & Francis, 2018) Fuentes, David; Muñoz, Nina M.; Guo, Chunxiao; Polak, Urzsula; Minhaj, Adeeb A.; Allen, William J.; Gustin, Michael C.; Cressman, Erik N.K.OBJECTIVE: A molecular dynamics approach to understanding fundamental mechanisms of combined thermal and osmotic stress induced by thermochemical ablation (TCA) is presented. METHODS: Structural models of fibronectin and fibronectin bound to its integrin receptor provide idealized models for the effects of thermal and osmotic stress in the extracellular matrix. Fibronectin binding to integrin is known to facilitate cell survival. The extracellular environment produced by TCA at the lesion boundary was modelled at 37 °C and 43 °C with added sodium chloride (NaCl) concentrations (0, 40, 80, 160, and 320 mM). Atomistic simulations of solvated proteins were performed using the GROMOS96 force field and TIP3P water model. Computational results were compared with the results of viability studies of human hepatocellular carcinoma (HCC) cell lines HepG2 and Hep3B under matching thermal and osmotic experimental conditions. RESULTS: Cell viability was inversely correlated with hyperthermal and hyperosmotic stresses. Added NaCl concentrations were correlated with a root mean square fluctuation increase of the fibronectin arginylglycylaspartic acid (RGD) binding domain. Computed interaction coefficients demonstrate preferential hydration of the protein model and are correlated with salt-induced strengthening of hydrophobic interactions. Under the combined hyperthermal and hyperosmotic stress conditions (43 °C and 320 mM added NaCl), the free energy change required for fibronectin binding to integrin was less favorable than that for binding under control conditions (37 °C and 0 mM added NaCl). CONCLUSION: Results quantify multiple measures of structural changes as a function of temperature increase and addition of NaCl to the solution. Correlations between cell viability and stability measures suggest that protein aggregates, non-functional proteins, and less favorable cell attachment conditions have a role in TCA-induced cell stress.Item Interactive Brain Tumor Segmentation(2023-06-08) Balsells, Cito; Riviere, Beatrice; Fuentes, DavidMachine learning based image segmentation relies on having access to a large dataset of labeled scans. Challenges arise when a sufficient training dataset is not available. To build a labeled dataset, one can manually create segmentations either by hand or assisted by a semi-automatic interactive tool. Here, interaction is given by the user in form of foreground and background clicks on the image. This thesis evaluates semi-automatic interactive image segmentation models applied to 3D brain tumor segmentation from MRI scans under two conditions. The first condition involves training models on the entire dataset. This provides a baseline for the best expected performance for each model. The second condition involves training models on portions of the dataset. This is done in an effort to model a dataset being gradually created from a small dataset. In both conditions, we find that the Wilcoxon signed rank test indicates significant results when comparing some of our interactive models with their fully-automatic counterpart. However, we ultimately deem the difference to be clinically irrelevant.Item Machine Learning Methods for Vessel Segmentation in Organs(2022-04-19) Tzolova, Bilyana; Riviere, Beatrice; Fuentes, DavidThe vascular system plays a crucial role in diagnostics, treatment, and surgical planning in a wide array of diseases. Recently, there has been a growing interest in automating the manual vessel segmentation process to save time. We aim to efficiently and effectively segment the vascular system in the liver organ using deep learning techniques in order to improve on current manual methods. We propose a 3D DenseNet using PocketNet paradigm with binary and ternary classifications that has less parameters to train than the state of the art methods. We explore the impact of various preprocessing techniques on the accuracy of the neural network. We are able to reduce training times and increase accuracy per training parameter in medical imaging segmentation of the liver vessels. Finally, we assess the accuracy of our model predictions using the dice score coefficient. We find that successful preprocessing filters and neural network parameters are necessary for consistently high dice scores.Item PocketNet: A Smaller Neural Network For Medical Image Analysis(2023-04-24) Celaya, Adrian; Riviere, Beatrice; Fuentes, DavidMedical imaging deep learning models are often large and complex, requiring specialized hardware to train and evaluate these models. To address such issues, we propose the PocketNet paradigm to reduce the size of deep learning models by throttling the growth of the number of channels in convolutional neural networks. We demonstrate that, for a range of segmentation and classification tasks, PocketNet architectures produce results comparable to that of conventional neural networks while reducing the number of parameters by multiple orders of magnitude, using up to 90% less GPU memory, and speeding up training times by up to 40%, thereby allowing such models to be trained and deployed in resource-constrained settings.