Semantic segmentation of plant roots from RGB (mini-) rhizotron images-generalisation potential and false positives of established methods and advanced deep-learning models

BACKGROUND

Manual analysis of (mini-)rhizotron (MR) images is tedious. Several methods have been proposed for semantic root segmentation based on homogeneous, single-source MR datasets. Recent advances in deep learning (DL) have enabled automated feature extraction, but comparisons of segmentation accuracy, false positives and transferability are virtually lacking. Here we compare six state-of-the-art methods and propose two improved DL models for semantic root segmentation using a large MR dataset with and without augmented data. We determine the performance of the methods on a homogeneous maize dataset, and a mixed dataset of > 8 species (mixtures), 6 soil types and 4 imaging systems. The generalisation potential of the derived DL models is determined on a distinct, unseen dataset.

RESULTS

The best performance was achieved by the U-Net models; the more complex the encoder the better the accuracy and generalisation of the model. The heterogeneous mixed MR dataset was a particularly challenging for the non-U-Net techniques. Data augmentation enhanced model performance. We demonstrated the improved performance of deep meta-architectures and feature extractors, and a reduction in the number of false positives.

CONCLUSIONS

Although correction factors are still required to match human labelled root lengths, neural network architectures greatly reduce the time required to compute the root length. The more complex architectures illustrate how future improvements in root segmentation within MR images can be achieved, particularly reaching higher segmentation accuracies and model generalisation when analysing real-world datasets with artefacts-limiting the need for model retraining.

Automatic segmentation of cardiac magnetic resonance images based on multi-input fusion network

PURPOSE

Using computer-assisted means to process a large amount of heart image data in order to speed up the diagnosis efficiency and accuracy of medical doctors has become a research worthy of investigation.

METHOD

Based on the U-Net model, this paper proposes a multi-input fusion network (MIFNet) model based on multi-scale input and feature fusion, which automatically extracts and fuses features of different input scales to realize the detection of Cardiac Magnetic Resonance Images (CMRI). The MIFNet model is trained and verified on the public data set, and then compared with the segmentation models, namely the Fully Convolutional Network (FCN) and DeepLab v1.

RESULTS

MIFNet model segmentation of CMRI significantly improved the segmentation accuracy, and the Dice value reached 97.238%. Compared with FCN and DeepLab v1, the average Hausdorff distance (HD) was reduced by 16.425%. The capacity parameter of FCN is 124.86% of MIFNet, DeepLab v1 is 103.22% of MIFNet.

CONCLUSION

Our proposed MIFNet model reduces the amount of parameters and improves the training speed while ensuring the simultaneous segmentation of overlapping targets. It can help clinicians to more quickly check the patient's CMRI focus area, and thereby improving the efficiency of diagnosis.

The role of computational methods for automating and improving clinical target volume definition

Treatment planning in radiotherapy distinguishes three target volume concepts: the gross tumor volume (GTV), the clinical target volume (CTV), and the planning target volume (PTV). Over time, GTV definition and PTV margins have improved through the development of novel imaging techniques and better image guidance, respectively. CTV definition is sometimes considered the weakest element in the planning process. CTV definition is particularly complex since the extension of microscopic disease cannot be seen using currently available in-vivo imaging techniques. Instead, CTV definition has to incorporate knowledge of the patterns of tumor progression. While CTV delineation has largely been considered the domain of radiation oncologists, this paper, arising from a 2019 ESTRO Physics research workshop, discusses the contributions that medical physics and computer science can make by developing computational methods to support CTV definition. First, we overview the role of image segmentation algorithms, which may in part automate CTV delineation through segmentation of lymph node stations or normal tissues representing anatomical boundaries of microscopic tumor progression. The recent success of deep convolutional neural networks has also enabled learning entire CTV delineations from examples. Second, we discuss the use of mathematical models of tumor progression for CTV definition, using as example the application of glioma growth models to facilitate GTV-to-CTV expansion for glioblastoma that is consistent with neuroanatomy. We further consider statistical machine learning models to quantify lymphatic metastatic progression of tumors, which may eventually improve elective CTV definition. Lastly, we discuss approaches to incorporate uncertainty in CTV definition into treatment plan optimization as well as general limitations of the CTV concept in the case of infiltrating tumors without natural boundaries.

Comparison of Automatic Segmentation Algorithms for the Subthalamic Nucleus

INTRODUCTION

Various automatic segmentation algorithms for the subthalamic nucleus (STN) have been published recently. However, most of the available software tools are not approved for clinical use.

OBJECTIVE

The aim of this study is to evaluate a clinically available automatic segmentation tool of the navigation planning software Brainlab Elements (BL-E) by comparing the output to manual segmentation and a nonclinically approved research method using the DISTAL atlas (DA) and the Horn electrophysiological atlas (HEA).

METHODS

Preoperative MRI data of 30 patients with idiopathic Parkinson's disease were used, resulting in 60 STN segmentations. The segmentations were created manually by two clinical experts. Automatic segmentations of the STN were obtained from BL-E and Advanced Normalization Tools using DA and HEA. Differences between manual and automatic segmentations were quantified by Dice and Jaccard coefficient, target overlap, and false negative/positive value (FNV/FPV) measurements. Statistical differences between similarity measures were assessed using the Wilcoxon signed-rank test with continuity correction, and comparison with interrater results was performed using the Mann-Whitney U test.

RESULTS

For manual segmentation, the mean size of the segmented STN was 133 ± 24 mm3. The mean size of the STN was 121 ± 18 mm3 for BL-E, 162 ± 21 mm3 for DA, and 130 ± 17 mm3 for HEA. The Dice coefficient for the interrater comparison was 0.63 and 0.54 ± 0.12, 0.59 ± 0.13, and 0.52 ± 0.14 for BL-E, DA, and HEA, respectively. Significant differences between similarity measures were found for Dice and Jaccard coefficient, target overlap and FNV between BL-E and DA; and FPV between BL-E and HEA. However, none of the differences were significant compared to interrater variability. The analysis of the center of gravity of the segmentations revealed that the BL-E STN ROI was located more medially, superior and posterior compared to other segmentations. Regarding the target overlap for beta power within the STN ROI included with the HEA, the BL-E segmentation showed a significantly higher value compared to manual segmentation.

CONCLUSION

Automatic image segmentation by means of the clinically approved software BL-E provides STN segmentations with similar accuracy like research tools, and differences are in the range of observed interrater variability. Further studies are required to investigate the clinical validity, for example, by comparing segmentation results of BL-E with electrophysiological data.

Automatic initialization and quality control of large-scale cardiac MRI segmentations

Continuous advances in imaging technologies enable ever more comprehensive phenotyping of human anatomy and physiology. Concomitant reduction of imaging costs has resulted in widespread use of imaging in large clinical trials and population imaging studies. Magnetic Resonance Imaging (MRI), in particular, offers one-stop-shop multidimensional biomarkers of cardiovascular physiology and pathology. A wide range of analysis methods offer sophisticated cardiac image assessment and quantification for clinical and research studies. However, most methods have only been evaluated on relatively small databases often not accessible for open and fair benchmarking. Consequently, published performance indices are not directly comparable across studies and their translation and scalability to large clinical trials or population imaging cohorts is uncertain. Most existing techniques still rely on considerable manual intervention for the initialization and quality control of the segmentation process, becoming prohibitive when dealing with thousands of images. The contributions of this paper are three-fold. First, we propose a fully automatic method for initializing cardiac MRI segmentation, by using image features and random forests regression to predict an initial position of the heart and key anatomical landmarks in an MRI volume. In processing a full imaging database, the technique predicts the optimal corrective displacements and positions in relation to the initial rough intersections of the long and short axis images. Second, we introduce for the first time a quality control measure capable of identifying incorrect cardiac segmentations with no visual assessment. The method uses statistical, pattern and fractal descriptors in a random forest classifier to detect failures to be corrected or removed from subsequent statistical analysis. Finally, we validate these new techniques within a full pipeline for cardiac segmentation applicable to large-scale cardiac MRI databases. The results obtained based on over 1200 cases from the Cardiac Atlas Project show the promise of fully automatic initialization and quality control for population studies.

Automatic segmentation of abdominal organs and adipose tissue compartments in water-fat MRI: Application to weight-loss in obesity

PURPOSE

To develop a fully automatic algorithm for abdominal organs and adipose tissue compartments segmentation and to assess organ and adipose tissue volume changes in longitudinal water-fat magnetic resonance imaging (MRI) data.

MATERIALS AND METHODS

Axial two-point Dixon images were acquired in 20 obese women (age range 24-65, BMI 34.9±3.8kg/m(2)) before and after a four-week calorie restriction. Abdominal organs, subcutaneous adipose tissue (SAT) compartments (abdominal, anterior, posterior), SAT regions along the feet-head direction and regional visceral adipose tissue (VAT) were assessed by a fully automatic algorithm using morphological operations and a multi-atlas-based segmentation method.

RESULTS

The accuracy of organ segmentation represented by Dice coefficients ranged from 0.672±0.155 for the pancreas to 0.943±0.023 for the liver. Abdominal SAT changes were significantly greater in the posterior than the anterior SAT compartment (-11.4%±5.1% versus -9.5%±6.3%, p<0.001). The loss of VAT that was not located around any organ (-16.1%±8.9%) was significantly greater than the loss of VAT 5cm around liver, left and right kidney, spleen, and pancreas (p<0.05).

CONCLUSION

The presented fully automatic algorithm showed good performance in abdominal adipose tissue and organ segmentation, and allowed the detection of SAT and VAT subcompartments changes during weight loss.

Performance of automatic image segmentation algorithms for calculating total lesion glycolysis for early response monitoring in non-small cell lung cancer patients during concomitant chemoradiotherapy

BACKGROUND AND PURPOSE

This study evaluated the use of total lesion glycolysis (TLG) determined by different automatic segmentation algorithms, for early response monitoring in non-small cell lung cancer (NSCLC) patients during concomitant chemoradiotherapy.

MATERIALS AND METHODS

Twenty-seven patients with locally advanced NSCLC treated with concomitant chemoradiotherapy underwent (18)F-fluorodeoxyglucose (FDG) PET/CT imaging before and in the second week of treatment. Segmentation of the primary tumours and lymph nodes was performed using fixed threshold segmentation at (i) 40% SUVmax (T40), (ii) 50% SUVmax (T50), (iii) relative-threshold-level (RTL), (iv) signal-to-background ratio (SBR), and (v) fuzzy locally adaptive Bayesian (FLAB) segmentation. Association of primary tumour TLG (TLGT), lymph node TLG (TLGLN), summed TLG (TLGS=TLGT+TLGLN), and relative TLG decrease (ΔTLG) with overall-survival (OS) and progression-free survival (PFS) was determined using univariate Cox regression models.

RESULTS

Pretreatment TLGT was predictive for PFS and OS, irrespective of the segmentation method used. Inclusion of TLGLN improved disease and early response assessment, with pretreatment TLGS more strongly associated with PFS and OS than TLGT for all segmentation algorithms. This was also the case for ΔTLGS, which was significantly associated with PFS and OS, with the exception of RTL and T40.

CONCLUSIONS

ΔTLGS was significantly associated with PFS and OS, except for RTL and T40. Inclusion of TLGLN improves early treatment response monitoring during concomitant chemoradiotherapy with FDG-PET.

A Fast Method for the Segmentation of Synaptic Junctions and Mitochondria in Serial Electron Microscopic Images of the Brain

Recent electron microscopy (EM) imaging techniques permit the automatic acquisition of a large number of serial sections from brain samples. Manual segmentation of these images is tedious, time-consuming and requires a high degree of user expertise. Therefore, there is considerable interest in developing automatic segmentation methods. However, currently available methods are computationally demanding in terms of computer time and memory usage, and to work properly many of them require image stacks to be isotropic, that is, voxels must have the same size in the X, Y and Z axes. We present a method that works with anisotropic voxels and that is computationally efficient allowing the segmentation of large image stacks. Our approach involves anisotropy-aware regularization via conditional random field inference and surface smoothing techniques to improve the segmentation and visualization. We have focused on the segmentation of mitochondria and synaptic junctions in EM stacks from the cerebral cortex, and have compared the results to those obtained by other methods. Our method is faster than other methods with similar segmentation results. Our image regularization procedure introduces high-level knowledge about the structure of labels. We have also reduced memory requirements with the introduction of energy optimization in overlapping partitions, which permits the regularization of very large image stacks. Finally, the surface smoothing step improves the appearance of three-dimensional renderings of the segmented volumes.

Quantitative measurement of transcription dynamics in living cells

In a wide range of organisms the kinetics of transcription have been found to be noisy, with "bursts" or "pulses" of transcription interspersed with irregular periods of inactivity. The in vivo analysis of transcription dynamics can be most directly monitored using RNA stem loop motifs derived from MS2 and other bacteriophages. Here we describe the implementation of the MS2 RNA detection system and the steps required for precise measurement of transcription dynamics in highly motile cells. Automated image processing techniques are used to track large numbers of cells and measure transcription in a systematic and unbiased manner. We discuss popular methods for automatic image segmentation and frame-to-frame tracking of cells, and the considerations required to make measurements as quantitatively as possible.