Search

The human brain is a complex system with many functional units interacting with each other. This interacting relationship, known as the Functional Connectivity Network (FCN), is critical for brain functions. To learn the FCN, machine learning algorithms can be built based on brain signals captured by sensing technologies such as EEG and fMRI. In neurological diseases, past research has revealed that the FCN is altered. Also, focusing on a specific disease, some part of the FCN, i.e., a sub-network can be more susceptible than other parts. However, the current knowledge about disease-specific sub-networks is limited. We propose a novel Discriminant Subgraph Learner (DSL) to identify a functional sub-network that best differentiates patients with a specific disease from healthy controls based on brain sensory data. We develop an integrated optimization framework for DSL to simultaneously learn the FCN of each class and identify the discriminant sub-network. Further, we develop tractable and converging algorithms to solve the optimization. We apply DSL to identify a functional sub-network that best differentiates patients with episodic migraine from healthy controls based on a fMRI dataset. DSL achieved the best accuracy compared to five state-of-the-art competing algorithms. [ABSTRACT FROM AUTHOR]

Multimodality datasets are becoming increasingly common in various domains to provide complementary information for predictive analytics. One significant challenge in fusing multimodality data is that the multiple modalities are not universally available for all samples due to cost and accessibility constraints. This situation results in a unique data structure called an Incomplete Multimodality Dataset. 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. 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 Alzheimer's disease at an early stage called Mild Cognitive Impairment using incomplete multimodality imaging data. IMTL achieves higher accuracy than competing methods without transfer learning. [ABSTRACT FROM AUTHOR]

Advances in sensing technology have generated multi-modality datasets with complementary information in various domains. In health care, it is common to acquire images of different types/modalities for the same patient to facilitate clinical decision making. We propose a clustering method called hierarchically-structured Factor Mixture Model (hierFMM) that enables cluster discovery from multi-modality datasets to exploit their joint strength. HierFMM employs a novel double-L21-penalized likelihood formulation to achieve hierarchical selection of modalities and features that are nested within the modalities. This formulation is proven to satisfy a Quadratic Majorization condition that allows for an efficient Group-wise Majorization Descent algorithm to be developed for model estimation. Simulation studies show significantly better performance of hierFMM than competing methods. HierFMM is applied to an application of identifying clusters/subgroups of migraine patients based on brain cortical area, thickness, and volume datasets extracted from Magnetic Resonance Imaging. Two subgroups are found, whose patients significantly differ in clinical characteristics. This finding shows the promise of using multi-modality imaging data to help patient stratification and develop optimal treatment for different subgroups with migraine. [ABSTRACT FROM AUTHOR]

Radiation Therapy (RT) is one of the most common treatments for cancer. To understand the impact of radiation toxicity on normal tissue, a Normal Tissue Complication Probability (NTCP) model is needed to link RT dose with radiation-induced complications. There are two types of NTCP models: biological and statistical models. Biological models have good generalizability but low accuracy, as they cannot factor in patient-specific information. Statistical models can incorporate patient-specific variables, but may not generalize well across different studies. We propose an integrated model that borrows strength from both biological and statistical models. Specifically, we propose a novel model formulation followed by an efficient parameter estimation algorithm, and investigate statistical properties of the estimator. We apply the integrated model to a real dataset of prostate cancer patients treated with Intensity Modulated RT at the Mayo Clinic Arizona, who are at risk of developing the grade 2+ acute rectal complication. The integrated model achieves an Area Under the Curve (AUC) level of 0.82 in prediction, whereas the AUCs for the biological and statistical models are only 0.66 and 0.76, respectively. The superior performance of the integrated model is also consistently observed over different simulation experiments. [ABSTRACT FROM AUTHOR]