Search

Automated anomaly detection from medical images, such as MRIs and X-rays, can significantly reduce human effort in disease diagnosis. Owing to the complexity of modeling anomalies and the high cost of manual annotation by domain experts (e.g., radiologists), a typical technique in the current medical imaging literature has focused on deriving diagnostic models from healthy subjects only, assuming the model will detect the images from patients as outliers. However, in many real-world scenarios, unannotated datasets with a mix of both healthy and diseased individuals are abundant. Therefore, this paper poses the research question of how to improve unsupervised anomaly detection by utilizing (1) an unannotated set of mixed images, in addition to (2) the set of healthy images as being used in the literature. To answer the question, we propose HealthyGAN, a novel one-directional image-to-image translation method, which learns to translate the images from the mixed dataset to only healthy images. Being one-directional, HealthyGAN relaxes the requirement of cycle consistency of existing unpaired image-to-image translation methods, which is unattainable with mixed unannotated data. Once the translation is learned, we generate a difference map for any given image by subtracting its translated output. Regions of significant responses in the difference map correspond to potential anomalies (if any). Our HealthyGAN outperforms the conventional state-of-the-art methods by significant margins on two publicly available datasets: COVID-19 and NIH ChestX-ray14, and one institutional dataset collected from Mayo Clinic. The implementation is publicly available at https://github.com/mahfuzmohammad/HealthyGAN., Comment: International Workshop on Simulation and Synthesis in Medical Imaging, MICCAI, 2022

Multimodality datasets are becoming increasingly common in various domains to provide complementary information for predictive analytics. In health care, diagnostic imaging of different kinds contains complementary information about an organ of interest, which allows for building a predictive model to accurately detect a certain disease. In manufacturing, multi-sensory datasets contain complementary information about the process and product, allowing for more accurate quality assessment. One significant challenge in fusing multimodality data for predictive analytics is that the multiple modalities are not universally available for all samples due to cost and accessibility constraints. This results in a unique data structure called Incomplete Multimodality Dataset (IMD) for which existing statistical models fall short. We propose a novel Incomplete-Multimodality Transfer Learning (IMTL) model that builds a predictive model for each sub-cohort of samples with the same missing modality pattern, and meanwhile couples the model estimation processes for different sub-cohorts to allow for transfer learning. We develop an Expectation-Maximization (EM) algorithm to estimate the parameters of IMTL and further extend it to a collaborative learning paradigm that is specifically valuable for patient privacy preservation in health care applications of the IMTL. We prove two advantageous properties of IMTL: the ability for out-of-sample prediction and a theoretical guarantee for a larger Fisher information compared with models without transfer learning. IMTL is applied to diagnosis and prognosis of the Alzheimer’s Disease (AD) at an early stage of the disease called Mild Cognitive Impairment (MCI) using incomplete multimodality imaging data. IMTL achieves higher accuracy than competing methods without transfer learning.

Compressed sensing (CS) has become a popular field in the last two decades to represent and reconstruct a sparse signal with much fewer samples than the signal itself. Although regular images are not sparse on their own, many can be sparsely represented in wavelet transform domain. Therefore, CS has also been widely applied to represent digital images. However, an alternative approach, adaptive sampling such as mesh-based image representation (MbIR), has not attracted as much attention. MbIR works directly on image pixels and represents the image with fewer points using a triangular mesh. In this paper, we perform a preliminary comparison between the CS and a recently developed MbIR method, AMA representation. The results demonstrate that, at the same sample density, AMA representation can provide better reconstruction quality than CS based on the tested algorithms. Further investigation with recent algorithms is needed to perform a thorough comparison., Comment: 9 pages, 4 figures, 2 tables

EMSA of VarR/25bp varBC IR DNA complex with increasing concentrations of erythromycin. Lane 1 0.08ng 25bp varBC IR DNA only. Lanes 2 to 16 50ng VarR/0.08ng 25 bp varBC IR DNA complex with increasing titrations erythromycin (0, 0.5, 1, 2, 4, 8, 16, 32, 64, 128, 256,512, 1024ng, 10, 100ug, respectively).

For data-driven building energy forecasting modeling, the quality of training data strongly affects a model’s accuracy and cost-effectiveness. In order to obtain high-quality training data within a short time period, experiment design, active learning, or excitation is becoming increasingly important, especially for nonlinear systems such as building energy systems. Experiment design and system excitation have been widely studied and applied in fields such as robotics and automobile industry for their model development. But these methods have hardly been applied for building energy modeling. This paper presents an overall discussion on the topic of applying system excitation for developing building energy forecasting models. For gray-box and white-box models, a model’s physical representations and theories can be applied to guide their training data collections. However, for black-box (pure-data-driven) models, the training data’s quality is sensitive to the model structure, leading to a fact that there is no universal theory for data training.  The focus of black-box modeling has traditionally been on how to represent a data set well. The impact of how such a data set represents the real system and how the quality of a training data set affect the performances of black-box models have not been well studied. In this paper, the system excitation method, which is used in system identification area, is used to excite zone temperature set-points to generate training data. These training data from system excitation are then used to train a variety of black-box building energy forecasting models. The models’ performances (accuracy and extendibility) are compared among different model structures. For the same model structure, its performances are also compared between when it is trained using typical building operational data and when it is trained using exited training data. Results show that the black-box models trained by normal operation data achieve better performance than that trained by excited training data but have worse model extendibility; Training data obtained from excitation will help to improve performances of system identification models.

Naval Postgraduate School Acquisition Research Program

Tenth Annual Acquisition Research Symposium Acquisition Management Excerpt from the Proceedings of the Tenth Annual Acquisition Research Symposium Acquisition Management Disclaimer: The views represented in this report are those of the authors and do not reflect the official policy position of the Navy, the Department of Defense, or the federal government. Naval Postgraduate School Acquisition Research Program Prepared for the Naval Postgraduate School, Monterey, CA Naval Postgraduate School Acquisition Research Program Approved for public release; distribution is unlimited.