Microbiome-based disease prediction with multimodal variational information bottlenecks.

Scientific research is shedding light on the interaction of the gut microbiome with the human host and on its role in human health. Existing machine learning methods have shown great potential in discriminating healthy from diseased microbiome states. Most of them leverage shotgun metagenomic sequencing to extract gut microbial species-relative abundances or strain-level markers. Each of these gut microbial profiling modalities showed diagnostic potential when tested separately; however, no existing approach combines them in a single predictive framework. Here, we propose the Multimodal Variational Information Bottleneck (MVIB), a novel deep learning model capable of learning a joint representation of multiple heterogeneous data modalities. MVIB achieves competitive classification performance while being faster than existing methods. Additionally, MVIB offers interpretable results. Our model adopts an information theoretic interpretation of deep neural networks and computes a joint stochastic encoding of different input data modalities. We use MVIB to predict whether human hosts are affected by a certain disease by jointly analysing gut microbial species-relative abundances and strain-level markers. MVIB is evaluated on human gut metagenomic samples from 11 publicly available disease cohorts covering 6 different diseases. We achieve high performance (0.80 < ROC AUC < 0.95) on 5 cohorts and at least medium performance on the remaining ones. We adopt a saliency technique to interpret the output of MVIB and identify the most relevant microbial species and strain-level markers to the model's predictions. We also perform cross-study generalisation experiments, where we train and test MVIB on different cohorts of the same disease, and overall we achieve comparable results to the baseline approach, i.e. the Random Forest. Further, we evaluate our model by adding metabolomic data derived from mass spectrometry as a third input modality. Our method is scalable with respect to input data modalities and has an average training time of < 1.4 seconds. The source code and the datasets used in this work are publicly available.

Deep Convolution Neural Network (DCNN) Multiplane Approach to Synthetic CT Generation From MR images-Application in Brain Proton Therapy.

PURPOSE: The first aim of this work is to present a novel deep convolution neural network (DCNN) multiplane approach and compare it to single-plane prediction of synthetic computed tomography (sCT) by using the real computed tomography (CT) as ground truth. The second aim is to demonstrate the feasibility of magnetic resonance imaging (MRI)-based proton therapy planning for the brain by assessing the range shift error within the clinical acceptance threshold. METHODS AND MATERIALS: The image database included 15 pairs of MRI/CT scans of the head. Three DCNNs were trained to estimate, for each voxel, the Hounsfield unit (HU) value from MRI intensities. Each DCNN gave an estimation in the axial, sagittal, and coronal plane, respectively. The median HU among the 3 values was selected to build the sCT. The sCT/CT agreement was evaluated by a mean absolute error (MAE) and mean error, computed within the head contour and on 6 different tissues. Dice similarity coefficients were calculated to assess the geometric overlap of bone and air cavities segmentations. A 3-beam proton therapy plan was simulated for each patient. Beam-by-beam range shift (RS) analysis was conducted to assess the proton-stopping power estimation. RS analysis was performed using clinically accepted thresholds of (1) 3.5% + 1 mm and (2) 2.5% + 1.5 mm of the total range. RESULTS: DCNN multiplane statistically outperformed single-plane prediction of sCT (P < .025). MAE and mean error within the head were 54 ± 7 HU and -4 ± 17 HU (mean ± standard deviation), respectively. Soft tissues were very close to perfect agreement (11 ± 3 HU in terms of MAE). Segmentation of air and bone regions led to a Dice similarity coefficient of 0.92 ± 0.03 and 0.93 ± 0.02, respectively. Proton RS was always below clinical acceptance thresholds, with a relative RS error of 0.14% ± 1.11%. CONCLUSIONS: The multiplane DCNN approach significantly improved the sCT prediction compared with other DCNN methods presented in the literature. The method was demonstrated to be highly accurate for MRI-only proton planning purposes.

Advanced Multimodal Methods for Cranial Pseudo-CT Generation Validated by IMRT and VMAT Radiation Therapy Plans.

PURPOSE: To investigate advanced multimodal methods for pseudo-computed tomography (CT) generation from standard magnetic resonance imaging sequences and to validate the results by intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) plans. We present 2 novel methods that employ key techniques to enhance pseudo-CTs and investigate the effect on image quality and applicability for IMRT and VMAT planning. MATERIALS AND METHODS: The data set contains CT and magnetic resonance image scans from 15 patients who underwent cranial radiation therapy. For each patient, pseudo-CTs of the head were generated with a patch-based and a voxel-based algorithm. The accuracy of the pseudo-CTs in comparison to clinical CTs was evaluated by mean absolute error, bias, and the Dice coefficient (of bone). IMRT and VMAT plans were created for each patient. Dose distributions were calculated with both the pseudo-CT and the clinical CT scans and compared by gamma tests, dose-volume histograms, and isocenter doses. RESULTS: The generated pseudo-CTs exhibited average mean absolute errors of 118.7 ± 10.4 HU for the voxel-based algorithm and 73.0 ± 6.4 HU for the patch-based algorithm. The dose calculations were in good agreement and showed gamma test (2 mm, 2%) pass rates for both beam setups (IMRT and VMAT) of over 99% for 14 patients and over 98% for 1 patient. CONCLUSIONS: We showed that the key techniques of our 2 novel algorithms advance the quality of pseudo-CT significantly and generate very competitive pseudo-CTs compared with previously published methods. This quality was confirmed by low dose error in comparison to the ground-truth CT. With the achieved level of accuracy, our patch-based algorithm especially is a candidate for clinical routine use in IMRT and VMAT planning.

Proton range shift analysis on brain pseudo-CT generated from T1 and T2 MR.

BACKGROUND: In radiotherapy, MR imaging is only used because it has significantly better soft tissue contrast than CT, but it lacks electron density information needed for dose calculation. This work assesses the feasibility of using pseudo-CT (pCT) generated from T1w/T2w MR for proton treatment planning, where proton range comparisons are performed between standard CT and pCT. MATERIAL AND METHODS: MR and CT data from 14 glioblastoma patients were used in this study. The pCT was generated by using conversion libraries obtained from tissue segmentation and anatomical regioning of the T1w/T2w MR. For each patient, a plan consisting of three 18 Gy beams was designed on the pCT, for a total of 42 analyzed beams. The plan was then transferred onto the CT that represented the ground truth. Range shift (RS) between pCT and CT was computed at R80 over 10 slices. The acceptance threshold for RS was according to clinical guidelines of two institutions. A γ-index test was also performed on the total dose for each patient. RESULTS: Mean absolute error and bias for the pCT were 124 ± 10 and -16 ± 26 Hounsfield Units (HU), respectively. The median and interquartile range of RS was 0.5 and 1.4 mm, with highest absolute value being 4.4 mm. Of the 42 beams, 40 showed RS less than the clinical range margin. The two beams with larger RS were both in the cranio-caudal direction and had segmentation errors due to the partial volume effect, leading to misassignment of the HU. CONCLUSIONS: This study showed the feasibility of using T1w and T2w MRI to generate a pCT for proton therapy treatment, thus avoiding the use of a planning CT and allowing better target definition and possibilities for online adaptive therapies. Further improvements of the methodology are still required to improve the conversion from MRI intensities to HUs.

Longitudinal Motion Assessment of the Carotid Artery Using Speckle Tracking and Scale-Invariant Feature Transform.

The purpose of this work is to present and validate a novel approach for ultra-sound-based speckle tracking to measure the carotid artery longitudinal displacement, and to assess the apparent sliding between of Intima-Media Complex (IMC) and Adventitia (Ad) layers. This method utilizes feature detectors and descriptors to localize and track keypoints for local motion quantification. The procedure was tested and validated on an in silico dataset and on 18 heathy volunteers and 16 patients. Accuracy measured on in silico data gave a mean ± standard deviation of 23 ± 15 and 19 ± 18 µm for IMC and Ad respectively, and thus smaller than the pixel size (0.0925 mm). Robustness analysis was performed on in vivo images, obtaining a maximum variation coefficient, over 5 repeated measures, of 9.5 and 13.8% for IMC and Ad, respectively. The novel method capability for detecting the relative motion of IMC vs. Ad was compared with visual assessment performed by 2 physicians, leading to a correlation coefficient R of 0.7 in the worst case. (Healthy group scored by rater #1.) In conclusion, our results provide evidence that the novel method is able to accurately and reliably track carotid artery layer motion and that it overcomes limitations currently present in the literature, therefore providing an automatic tool for clinical evaluation of IMC vs. Ad relative displacement.