This thesis investigates the performance of deep learning models, specifically Resnet50 and TransUnet, in semantic image segmentation on microwell images containing tumor and natural killer (NK) cells. The main goal is to examine the effect of only using bright-field data (1-channel) as input instead of both fluorescent and brightfield data (4-channel); this is interesting since fluorescent imaging can cause damage to the cells being analyzed. The network performance is measured by Intersection over Union (IoU), the networks were trained and using manually annotated data from Onfelt Lab. TransUnet consistently outperformed the Resnet50 for both the 4-channel and 1-channel data. Moreover, the 4-channel input generally resulted in a better IoU compared to using only the bright-field channel. Furthermore, a significant decline in performance is observed when the networks are tested on the control data. For the control data, the overall IoU for the best performing 4-channel model dropped from 86.2\% to 73.9\%. The best performing 1-channel model dropped from 83.8\% to 70.8\% overall IoU.

abstract: Discriminative learning when training and test data belong to different distributions is a challenging and complex task. Often times we have very few or no labeled data from the test or target distribution, but we may have plenty of labeled data from one or multiple related sources with different distributions. Due to its capability of migrating knowledge from related domains, transfer learning has shown to be effective for cross-domain learning problems. In this dissertation, I carry out research along this direction with a particular focus on designing efficient and effective algorithms for BioImaging and Bilingual applications. Specifically, I propose deep transfer learning algorithms which combine transfer learning and deep learning to improve image annotation performance. Firstly, I propose to generate the deep features for the Drosophila embryo images via pretrained deep models and build linear classifiers on top of the deep features. Secondly, I propose to fine-tune the pretrained model with a small amount of labeled images. The time complexity and performance of deep transfer learning methodologies are investigated. Promising results have demonstrated the knowledge transfer ability of proposed deep transfer algorithms. Moreover, I propose a novel Robust Principal Component Analysis (RPCA) approach to process the noisy images in advance. In addition, I also present a two-stage re-weighting framework for general domain adaptation problems. The distribution of source domain is mapped towards the target domain in the first stage, and an adaptive learning model is proposed in the second stage to incorporate label information from the target domain if it is available. Then the proposed model is applied to tackle cross lingual spam detection problem at LinkedIn’s website. Our experimental results on real data demonstrate the efficiency and effectiveness of the proposed algorithms. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2015

Un crible biologique a pour objectif de tester en parallèle l'impact de nombreuses conditions expérimentales sur un processus biologique d'un organisme modèle. Le progrès technique et computationnel a rendu possible la réalisation de tels cribles à grande échelle - jusqu'à des centaines de milliers de conditions. L'imagerie sur cellules vivantes est un excellent outil pour étudier en détail les conséquences d'une perturbation chimique sur un processus biologique. L'analyse des cribles sur cellules vivantes demande toutefois la combinaison de méthodes robustes d'imagerie par ordinateur et de contrôle qualité, et d'approches statistiques efficaces pour la détection des effets significatifs. La présente thèse répond à ces défis par le développement de méthodes analytiques pour les images de cribles temporels à haut débit. Les cadres qui y sont développés sont appliqués à des données publiées, démontrant par là leur applicabilité ainsi que les bénéfices d'une ré-analyse des données de cribles à haut contenu (HCS). Le premier workflow pour l'étude de la motilité cellulaire à l'échelle d'une cellule dans de telles données constitue le chapitre 2. Le chapitre 3 applique ce workflow à des données publiées et présente une nouvelle distance pour l'inférence de cible thérapeutique à partir d'images de cribles temporels. Enfin, le chapitre 4 présente une pipeline méthodologique complète pour la conduite de cribles temporels à haut débit en toxicologie environnementale. / Biological screens test large sets of experimental conditions with respect to their specific biological effect on living systems. Technical and computational progresses have made it possible to perform such screens at a large scale - up to hundreds of thousands of experiments. Live cell imaging is an excellent tool to study in detail the consequences of chemical perturbation on a given biological process. However, the analysis of live cell screens demands the combination of robust computer vision methods, efficient statistical methods for the detection of significant effects and robust procedures for quality control. This thesis addresses these challenges by developing analytical methods for the analysis of High Throughput time-lapse microscopy screening data. The developed frameworks are applied to publicly available HCS data, demonstrating their applicability and the benefits of HCS data remining. The first multivariate workflow for the study of single cell motility in such large-scale data is detailed in Chapter 2. Chapter 3 presents this workflow application to previously published data, and the development of a new distance for drug target inference by in silico comparisons of parallel siRNA and drug screens. Finally, chapter 4 presents a complete methodological pipeline for performing HT time-lapse screens in Environmental Toxicology.