Structural neuroimaging as clinical predictor: A review of machine learning applications.

In this paper, we provide an extensive overview of machine learning techniques applied to structural magnetic resonance imaging (MRI) data to obtain clinical classifiers. We specifically address practical problems commonly encountered in the literature, with the aim of helping researchers improve the application of these techniques in future works. Additionally, we survey how these algorithms are applied to a wide range of diseases and disorders (e.g. Alzheimer's disease (AD), Parkinson's disease (PD), autism, multiple sclerosis, traumatic brain injury, etc.) in order to provide a comprehensive view of the state of the art in different fields.

Defining a multimodal signature of remote sports concussions.

Sports-related concussions lead to persistent anomalies of the brain structure and function that interact with the effects of normal ageing. Although post-mortem investigations have proposed a bio-signature of remote concussions, there is still no clear in vivo signature. In the current study, we characterized white matter integrity in retired athletes with a history of remote concussions by conducting a full-brain, diffusion-based connectivity analysis. Next, we combined MRI diffusion markers with MR spectroscopic, MRI volumetric, neurobehavioral and genetic markers to identify a multidimensional in vivo signature of remote concussions. Machine learning classifiers trained to detect remote concussions using this signature achieved detection accuracies up to 90% (sensitivity: 93%, specificity: 87%). These automated classifiers identified white matter integrity as the hallmark of remote concussions and could provide, following further validation, a preliminary unbiased detection tool to help medical and legal experts rule out concussion history in patients presenting or complaining about late-life abnormal cognitive decline.

The Disconnection Hypothesis in Alzheimer's Disease Studied Through Multimodal Magnetic Resonance Imaging: Structural, Perfusion, and Diffusion Tensor Imaging.

According to the so-called disconnection hypothesis, the loss of synaptic inputs from the medial temporal lobes (MTL) in Alzheimer's disease (AD) may lead to reduced activity of target neurons in cortical areas and, consequently, to decreased cerebral blood flow (CBF) in those areas. The aim of this study was to assess whether hypoperfusion in parietotemporal and frontal cortices of patients with mild cognitive impairment who converted to AD (MCI-c) and patients with mild AD is associated with atrophy in the MTL and/or microstructural changes in the white matter (WM) tracts connecting these areas. We assessed these relationships by investigating correlations between CBF in hypoperfused areas, mean cortical thickness in atrophied regions of the MTL, and fractional anisotropy (FA) in WM tracts. In the MCI-c group, a strong correlation was observed between CBF of the superior parietal gyri and FA in the parahippocampal tracts (left: r = 0.90, p <  0.0001; right: r = 0.597, p = 0.024), and between FA in the right parahippocampal tract and the right precuneus (r = 0.551, p = 0.041). No significant correlations between CBF in hypoperfused regions and FA in the WM tract were observed in the AD group. These results suggest an association between perfusion deficits and altered WM tracts in prodromal AD, while microvasculature impairments may have a greater influence in more advanced stages. We did not find correlations between cortical thinning in the medial temporal lobes and decreased FA in the WM tracts of the limbic system in either group.

Functional segmentation of dynamic PET studies: Open source implementation and validation of a leader-follower-based algorithm.

UNLABELLED: We present a novel segmentation algorithm for dynamic PET studies that groups pixels according to the similarity of their time-activity curves. METHODS: Sixteen mice bearing a human tumor cell line xenograft (CH-157MN) were imaged with three different (68)Ga-DOTA-peptides (DOTANOC, DOTATATE, DOTATOC) using a small animal PET-CT scanner. Regional activities (input function and tumor) were obtained after manual delineation of regions of interest over the image. The algorithm was implemented under the jClustering framework and used to extract the same regional activities as in the manual approach. The volume of distribution in the tumor was computed using the Logan linear method. A Kruskal-Wallis test was used to investigate significant differences between the manually and automatically obtained volumes of distribution. RESULTS: The algorithm successfully segmented all the studies. No significant differences were found for the same tracer across different segmentation methods. Manual delineation revealed significant differences between DOTANOC and the other two tracers (DOTANOC - DOTATATE, p=0.020; DOTANOC - DOTATOC, p=0.033). Similar differences were found using the leader-follower algorithm. CONCLUSION: An open implementation of a novel segmentation method for dynamic PET studies is presented and validated in rodent studies. It successfully replicated the manual results obtained in small-animal studies, thus making it a reliable substitute for this task and, potentially, for other dynamic segmentation procedures.

Meningiomas: a comparative study of 68Ga-DOTATOC, 68Ga-DOTANOC and 68Ga-DOTATATE for molecular imaging in mice.

PURPOSE: The goal of this study was to compare the tumor uptake kinetics and diagnostic value of three (68)Ga-DOTA-labeled somatostatin analogues ((68)Ga-DOTATOC, (68)Ga-DOTANOC, and (68)Ga-DOTATATE) using PET/CT in a murine model with subcutaneous meningioma xenografts. METHODS: The experiment was performed with 16 male NUDE NU/NU mice bearing xenografts of a human meningioma cell line (CH-157MN). (68)Ga-DOTATOC, (68)Ga-DOTANOC, and (68)Ga-DOTATATE were produced in a FASTLab automated platform. Imaging was performed on an Argus small-animal PET/CT scanner. The SUVmax of the liver and muscle, and the tumor-to-liver (T/L) and tumor-to-muscle (T/M) SUV ratios were computed. Kinetic analysis was performed using Logan graphical analysis for a two-tissue reversible compartmental model, and the volume of distribution (Vt) was determined. RESULTS: Hepatic SUVmax and Vt were significantly higher with (68)Ga-DOTANOC than with (68)Ga-DOTATOC and (68)Ga-DOTATATE. No significant differences between tracers were found for SUVmax in tumor or muscle. No differences were found in the T/L SUV ratio between (68)Ga-DOTATATE and (68)Ga-DOTATOC, both of which had a higher fraction than (68)Ga-DOTANOC. The T/M SUV ratio was significantly higher with (68)Ga-DOTATATE than with (68)Ga-DOTATOC and (68)Ga-DOTANOC. The Vt for tumor was higher with (68)Ga-DOTATATE than with (68)Ga-DOTANOC and relatively similar to that of (68)Ga-DOTATOC. CONCLUSIONS: This study demonstrates, for the first time, the ability of the three radiolabeled somatostatin analogues tested to image a human meningioma cell line. Although Vt was relatively similar with (68)Ga-DOTATATE and (68)Ga-DOTATOC, uptake was higher with (68)Ga-DOTATATE in the tumor than with (68)Ga-DOTANOC and (68)Ga-DOTATOC, suggesting a higher diagnostic value of (68)Ga-DOTATATE for detecting meningiomas.

jClustering, an open framework for the development of 4D clustering algorithms.

We present jClustering, an open framework for the design of clustering algorithms in dynamic medical imaging. We developed this tool because of the difficulty involved in manually segmenting dynamic PET images and the lack of availability of source code for published segmentation algorithms. Providing an easily extensible open tool encourages publication of source code to facilitate the process of comparing algorithms and provide interested third parties with the opportunity to review code. The internal structure of the framework allows an external developer to implement new algorithms easily and quickly, focusing only on the particulars of the method being implemented and not on image data handling and preprocessing. This tool has been coded in Java and is presented as an ImageJ plugin in order to take advantage of all the functionalities offered by this imaging analysis platform. Both binary packages and source code have been published, the latter under a free software license (GNU General Public License) to allow modification if necessary.

Is the cerebellum the optimal reference region for intensity normalization of perfusion MR studies in early Alzheimer's disease?

The cerebellum is the region most commonly used as a reference when normalizing the intensity of perfusion images acquired using magnetic resonance imaging (MRI) in Alzheimer's disease (AD) studies. In addition, the cerebellum provides unbiased estimations with nuclear medicine techniques. However, no reports confirm the cerebellum as an optimal reference region in MRI studies or evaluate the consequences of using different normalization regions. In this study, we address the effect of using the cerebellum, whole-brain white matter, and whole-brain cortical gray matter in the normalization of cerebral blood flow (CBF) parametric maps by comparing patients with stable mild cognitive impairment (MCI), patients with AD and healthy controls. According to our results, normalization by whole-brain cortical gray matter enables more sensitive detection of perfusion abnormalities in AD patients and reveals a larger number of affected regions than data normalized by the cerebellum or whole-brain white matter. Therefore, the cerebellum is not the most valid reference region in MRI studies for early stages of AD. After normalization by whole-brain cortical gray matter, we found a significant decrease in CBF in both parietal lobes and an increase in CBF in the right medial temporal lobe. We found no differences in perfusion between patients with stable MCI and healthy controls either before or after normalization.

Comparative evaluation of autofocus algorithms for a real-time system for automatic detection of Mycobacterium tuberculosis.

Microscopy images must be acquired at the optimal focal plane for the objects of interest in a scene. Although manual focusing is a standard task for a trained observer, automatic systems often fail to properly find the focal plane under different microscope imaging modalities such as bright field microscopy or phase contrast microscopy. This article assesses several autofocus algorithms applied in the study of fluorescence-labeled tuberculosis bacteria. The goal of this work was to find the optimal algorithm in order to build an automatic real-time system for diagnosing sputum smear samples, where both accuracy and computational time are important. We analyzed 13 focusing methods, ranging from well-known algorithms to the most recently proposed functions. We took into consideration criteria that are inherent to the autofocus function, such as accuracy, computational cost, and robustness to noise and to illumination changes. We also analyzed the additional benefit provided by preprocessing techniques based on morphological operators and image projection profiling.