Automatic segmentation of thoracic CT images using three deep learning models

PurposeDeep learning (DL) techniques are widely used in medical imaging and in particular for segmentation. Indeed, manual segmentation of organs at risk (OARs) is time-consuming and suffers from inter- and intra-observer segmentation variability. Image segmentation using DL has given very promising results. In this work, we present and compare the results of segmentation of OARs and a clinical target volume (CTV) in thoracic CT images using three DL models.Materials and methodsWe used CT images of 52 patients with breast cancer from a public dataset. Automatic segmentation of the lungs, the heart and a CTV was performed using three models based on the U-Net architecture. Three metrics were used to quantify and compare the segmentation results obtained with these models: the Dice similarity coefficient (DSC), the Jaccard coefficient (J) and the Hausdorff distance (HD).ResultsThe obtained values of DSC, J and HD were presented for each segmented organ and for the three models. Examples of automatic segmentation were presented and compared to the corresponding ground truth delineations. Our values were also compared to recent results obtained by other authors.ConclusionThe performance of three DL models was evaluated for the delineation of the lungs, the heart and a CTV. This study showed clearly that these 2D models based on the U-Net architecture can be used to delineate organs in CT images with a good performance compared to other models. Generally, the three models present similar performances. Using a dataset with more CT images, the three models should give better results.     

Automated Computational Detection of Interstitial Fibrosis,Tubular Atrophy,and Glomerulosclerosis

Open in a separate window     

Automated total kidney volume measurements in pre-clinical magnetic resonance imaging for resourcing imaging data,annotations, and source code

1. Download : Download high-res image (287KB)
2. Download : Download full-size image

     

Electrophysiology of statistical learning: Exploring the online learning process and offline learning product

A continuous stream of syllables is segmented into discrete constituents based on the transitional probabilities (TPs) between adjacent syllables by means of statistical learning. However, we still do not know whether people attend to high TPs between frequently co‐occurring syllables and cluster them together as parts of the discrete constituents or attend to low TPs aligned with the edges between the constituents and extract them as whole units. Earlier studies on TP‐based segmentation also have not distinguished between the segmentation process (how people segment continuous speech) and the learning product (what is learnt by means of statistical learning mechanisms). In the current study, we explored the learning outcome separately from the learning process, focusing on three possible learning products: holistic constituents that are retrieved from memory during the recognition test, clusters of frequently co‐occurring syllables, or a set of statistical regularities which can be used to reconstruct legitimate candidates for discrete constituents during the recognition test. Our data suggest that people employ boundary‐finding mechanisms during online segmentation by attending to low inter‐syllabic TPs during familiarization and also identify potential candidates for discrete constituents based on their statistical congruency with rules extracted during the learning process. Memory representations of recurrent constituents embedded in the continuous speech stream during familiarization facilitate subsequent recognition of these discrete constituents.     

Tractostorm: The what,why, and how of tractography dissection reproducibility

Investigative studies of white matter (WM) brain structures using diffusion MRI (dMRI) tractography frequently require manual WM bundle segmentation, often called “virtual dissection.” Human errors and personal decisions make these manual segmentations hard to reproduce, which have not yet been quantified by the dMRI community. It is our opinion that if the field of dMRI tractography wants to be taken seriously as a widespread clinical tool, it is imperative to harmonize WM bundle segmentations and develop protocols aimed to be used in clinical settings. The EADC‐ADNI Harmonized Hippocampal Protocol achieved such standardization through a series of steps that must be reproduced for every WM bundle. This article is an observation of the problematic. A specific bundle segmentation protocol was used in order to provide a real‐life example, but the contribution of this article is to discuss the need for reproducibility and standardized protocol, as for any measurement tool. This study required the participation of 11 experts and 13 nonexperts in neuroanatomy and “virtual dissection” across various laboratories and hospitals. Intra‐rater agreement (Dice score) was approximately 0.77, while inter‐rater was approximately 0.65. The protocol provided to participants was not necessarily optimal, but its design mimics, in essence, what will be required in future protocols. Reporting tractometry results such as average fractional anisotropy, volume or streamline count of a particular bundle without a sufficient reproducibility score could make the analysis and interpretations more difficult. Coordinated efforts by the diffusion MRI tractography community are needed to quantify and account for reproducibility of WM bundle extraction protocols in this era of open and collaborative science.     

Hippocampal state transitions at the boundaries between trial epochs

The hippocampus encodes distinct contexts with unique patterns of activity. Representational shifts with changes in context, referred to as remapping, have been extensively studied. However, less is known about transitions between representations. In this study, we leverage a large dataset of neuronal recordings taken while rats performed an olfactory memory task with a predictable temporal structure involving trials and intertrial intervals (ITIs), separated by salient boundaries at the trial start and trial end. We found that trial epochs were associated with stable hippocampal representations despite moment‐to‐moment variability in stimuli and behavior. Representations of trial and ITI epochs were far more distinct than spatial factors would predict and the transitions between the two were abrupt. The boundary was associated with a large spike in multiunit activity, with many individual cells specifically active at the start or end of each trial. Both epochs and boundaries were encoded by hippocampal populations, and these representations carried information on orthogonal axes readily identified using principal component analysis. We suggest that the hippocampus orthogonalizes representations of the trial and ITI epochs and the activity spike at trial boundaries might serve to drive hippocampal activity from one stable state to the other.     

Automated Delineation of the Normal Urinary Bladder on Planning CT and Cone Beam CT

PurposeTo quantify the volumetric effect of delineation variability when using manual versus semiautomated tools to contour the normal bladder on planning computed tomography (CT) and cone beam CT.MethodsFollowing research ethics board approval, 10 prostate cancer patients were selected. For each patient, one pretreatment cone beam CT (CBCT) was randomly selected from the first treatment week and registered to the planning CT (planCT). Model-based auto adaptation was used to delineate the outer bladder (OB) surface for the planCT. That contour was then propagated and manually adapted onto the CBCT. A second observer delineated OB for the planCT and CBCT using typical manual methods. These delineation procedures were repeated four times on each image set, with observers blinded to the previous contours. Metrics of volumetric, geometric, and overlap concordance were used to compare the manual and automated OB contours.ResultsThe mean pairwise difference between the manual and model-based planCT volumes was 4 cm³ (2%), and the model-based contours exhibited approximately half the observer variation of the manual ones (3 cm³, 2%). The mean of pairwise differences between the manual and propagated CBCT volumes was 13 cm³ (8%), but the propagated contours exhibited approximately half the observer related volume variation (11 cm³, 6%). Small CBCT bladder volumes displayed larger observer variation with manual methods (r², −0.640). Variability between the automated contours was significantly smaller than for the corresponding manual observations (P = .004 and .002, respectively). Metrics of three-dimensional overlap concordance indicated excellent agreement within and between the delineation methods. Automated CBCT contours were significantly smoother than the manual ones (surface sphericity index, 1.29 vs. 1.35; P = .03).ConclusionsVolumetric, geometric, and overlap metrics all indicated that planCT and CBCT automated OB contours fell within the range of manually delineated contours. The CBCT propagated contours were significantly smoother and associated with smaller intraobserver variability, compared with manual contours. Importantly, the findings from this research suggest that contour propagation may be more robust than manual delineation, especially in the presence of poor image quality.