Convolutional Neural Networks For Segmentation Research Articles

ACTIVITY CONCENTRATION ESTIMATION IN AUTOMATED KIDNEY SEGMENTATION BASED ON CONVOLUTION NEURAL NETWORK METHOD FOR 177LU-SPECT/CT KIDNEY DOSIMETRY.

For 177Lu-DOTATATE treatments, dosimetry based on manual kidney segmentation from computed tomography (CT) is accurate but time consuming and might be affected by misregistration between CT and SPECT images. This study develops a convolution neural network (CNN) for automated kidney segmentation that accurately aligns CT segmented volume of interest (VOI) to the kidneys in SPECT images. The CNN was trained with SPECT/CT images performed over the abdominal area of 137 patients treated with 177Lu-DOTATATE. Activity concentrations in automated and manual segmentations were strongly correlated for both kidneys (r>0.96, p<0.01) in the testing cohort (n=20). The Bland-Altman analyses demonstrated higher accuracy for the CNN segmentation compared to the manual segmented kidneys without VOI adjustment. The CNN demonstrated a potential for accurate kidney segmentation. The CNN was a fast and robust approach for assessment of activity concentrations in SPECT images, and performed equally well as the manual segmentation method.

Convolutional neural networks for segmentation of FIB-SEM nanotomography data from porous polymer films for controlled drug release.

Phase-separated polymer films are commonly used as coatings around pharmaceutical oral dosage forms (tablets or pellets) to facilitate controlled drug release. A typical choice is to use ethyl cellulose and hydroxypropyl cellulose (EC/HPC) polymer blends. When an EC/HPC film is in contact with water, the leaching out of the water-soluble HPC phase produces an EC film with a porous network through which the drug is transported. The drug release can be tailored by controlling the structure of this porous network. Imaging and characterization of such EC porous films facilitates understanding of how to control and tailor film formation and ultimately drug release. Combined focused ion beam and scanning electron microscope (FIB-SEM) tomography is a well-established technique for high-resolution imaging, and suitable for this application. However, for segmenting image data, in this case to correctly identify the porous network, FIB-SEM is a challenging technique to work with. In this work, we implement convolutional neural networks for segmentation of FIB-SEM image data. The data are acquired from three EC porous films where the HPC phases have been leached out. The three data sets have varying porosities in a range of interest for controlled drug release applications. We demonstrate very good agreement with manual segmentations. In particular, we demonstrate an improvement in comparison to previous work on the same data sets that utilized a random forest classifier trained on Gaussian scale-space features. Finally, we facilitate further development of FIB-SEM segmentation methods by making the data and software used openaccess.

A systematic evaluation of learning rate policies in training CNNs for brain tumor segmentation

Convolutional neural networks (CNNs) have recently emerged as a powerful approach for automatic segmentation of brain tumor subregions on 3D multi-parametric MRI scans. Learning rate is a crucial hyperparameter in the training of CNNs, impacting the performance of the learned model. Different learning rate policies trace unique trajectories in the optimization landscape that converge to local minima with varying generalization properties. In this work, we empirically evaluated nine learning rate policy-optimizer pairs with two state-of-the-art architectures, namely 2D slice-based U-Net and 3D DeepMedicRes, on an augmented brain tumor dataset of 534 subjects. Segmentation performance was quantified in terms of Dice similarity coefficient and Hausdorff distance metrics. The policies were ranked based on the final ranking score (FRS) employed by the BraTS challenge, with the statistical significance of the rankings evaluated by random permutation test. For 2D slice-based U-Net architecture, an overall ranking of learning rate policies showed that the polynomial decay policy with Adam optimizer significantly outperformed other policies for the task of individual and hierarchical segmentation of tumor subregions (p < 10−4). For 3D segment-based DeepMedicRes architecture, polynomial decay policy with Adam optimizer performed significantly better than all other policies, with the exception of polynomial decay with SGD optimizer for the same task (p < 10−4). Based on the FRS, polynomial decay policy with Adam and SGD optimizer occupied the top two positions respectively, but the difference was not statistically significant (p > 0.3). These findings were also validated on the BraTS 2019 Validation dataset which comprised of an additional 125 subjects.

Automated assessment of glomerulosclerosis and tubular atrophy using deep learning

In kidney transplantations, pathologists evaluate the architecture of both glomeruli, interstitium and tubules to assess the nephron status. An accurate assessment of glomerulosclerosis and tubular atrophy is crucial for determining kidney acceptance, which is currently based on the pathologists' histological evaluations on renal biopsies in addition to clinical data. In this work, we present an automated algorithm, called RENTAG (Robust EvaluatioN of Tubular Atrophy & Glomerulosclerosis), for the segmentation and classification of glomerular and tubular structures in histopathological images. The proposed novel strategy combines the accuracy of a level-set with the semantic segmentation of convolutional neural networks to detect the glomeruli and tubules contours. In the TEST set, our method exhibited excellent performance in both glomeruli (dice score: 0.9529) and tubule (dice score: 0.9174) detection and outperformed all the compared methods. To the best of our knowledge, the RENTAG algorithm is the first fully automated method capable of quantifying glomerulosclerosis and tubular atrophy in digital histological images. The developed software can be employed for the analysis of pre-transplantation biopsies to support the pathologists' diagnostic activity.

Multi-Muscle Deep Learning Segmentation to Automate the Quantification of Muscle Fat Infiltration in Cervical Spine Conditions

Muscle fat infiltration (MFI) is observed in participants with cervical spine disorders using magnetic resonance imaging. The quantification of MFI using magnetic resonance imaging (MRI) requires time-consuming and rater-dependent manual segmentation techniques. Convolutional neural network (CNN) models have demonstrated state-of-the-art performance in a single cervical spine muscle segmentation task in participants with whiplash injury following a motor vehicle collision (MVC). Here, we expanded this work and trained a CNN for multi-muscle segmentation and automatic MFI calculation of seven cervical spine muscle groups (left and right muscles segmented separately, 14 muscles total) from high-resolution fat-water imaging in participants following a MVC (34 training datasets, n = 17, 14 females, 3 males, age = 33.7 ± 11.4 years). First, we demonstrate high test reliability and accuracy of the CNN MFI measures compared to manual segmentation in an independent testing dataset (n = 18, 11 females, 7 males, age = 31.7 ± 9.6 years). Then in 84 participants imaged within two weeks following a MVC (61 females, 23 males, age = 34.2 ± 10.7 years), we examined MFI across the muscle groups and explored the relationship between MFI and sex, age, and body mass index (BMI). Averaging across all muscle groups, females had significantly higher MFI than males (p 0.300, p

Automatic segmentation of the left ventricle in echocardiographic images using convolutional neural networks.

Two-dimensional echocardiography (2D echo) is the most widely used non-invasive imaging modality due to its fast acquisition time, low cost, and high temporal resolution. Boundary identification of left ventricle (LV) in 2D echo, i.e., image segmentation, is the first step to calculate relevant clinical parameters. Currently, LV segmentation in 2D echo is primarily conducted semi-manually. A fully-automatic segmentation of the LV wall needs further development. We evaluated the performance of the state-of-the-art convolutional neural networks (CNNs) for the segmentation of 2D echo images from 6 standard projections of the LV. We used two segmentation algorithms: U-net and segAN. The models were trained using an in-house dataset, which consists of 1,649 porcine images from 6 to 8 different pigs. In addition, a transfer learning approach was used for the segmentation of long-axis projections by training models with our database based on the previously trained weights obtained from Cardiac Acquisitions for Multi-structure Ultrasound Segmentation (CAMUS) dataset. The models were tested on a separate set of images from two other pigs by computing several metrics. The segmentation process was combined with a 3D reconstruction framework to quantify the physiological indices such as LV volumes and ejection fraction (EF). The average dice metric for the LV cavity was 0.90 and 0.91 for the U-net and segAN, respectively, which was higher than 0.82 for the level-set (P value: 3.31×10-25). The average Hausdorff distance for the LV cavity was 2.71 mm and 2.82 mm for the U-net and segAN, respectively, which was lower than 3.64 mm for the level-set (P value: 4.86×10-16). The LV shapes and volumes obtained using the CNN segmentation models were in good agreement with the results segmented by the experts. In addition, the differences of the calculated physiological parameters between two 3D reconstruction models segmented by the experts and CNNs were less than 15%. The results showed that both CNN models achieve higher performance on LV segmentation than the level-set method. The error of the reconstruction from automatic segmentation compared to the expert segmentation is less than 15%, which is within the 20% error of echo compared to the gold standard.

A deep convolutional neural network for segmentation of whole-slide pathology images identifies novel tumour cell-perivascular niche interactions that are associated with poor survival in glioblastoma

BackgroundGlioblastoma is the most aggressive type of brain cancer with high-levels of intra- and inter-tumour heterogeneity that contribute to its rapid growth and invasion within the brain. However, a spatial characterisation of gene signatures and the cell types expressing these in different tumour locations is still lacking.MethodsWe have used a deep convolutional neural network (DCNN) as a semantic segmentation model to segment seven different tumour regions including leading edge (LE), infiltrating tumour (IT), cellular tumour (CT), cellular tumour microvascular proliferation (CTmvp), cellular tumour pseudopalisading region around necrosis (CTpan), cellular tumour perinecrotic zones (CTpnz) and cellular tumour necrosis (CTne) in digitised glioblastoma histopathological slides from The Cancer Genome Atlas (TCGA). Correlation analysis between segmentation results from tumour images together with matched RNA expression data was performed to identify genetic signatures that are specific to different tumour regions.ResultsWe found that spatially resolved gene signatures were strongly correlated with survival in patients with defined genetic mutations. Further in silico cell ontology analysis along with single-cell RNA sequencing data from resected glioblastoma tissue samples showed that these tumour regions had different gene signatures, whose expression was driven by different cell types in the regional tumour microenvironment. Our results further pointed to a key role for interactions between microglia/pericytes/monocytes and tumour cells that occur in the IT and CTmvp regions, which may contribute to poor patient survival.ConclusionsThis work identified key histopathological features that correlate with patient survival and detected spatially associated genetic signatures that contribute to tumour-stroma interactions and which should be investigated as new targets in glioblastoma. The source codes and datasets used are available in GitHub: https://github.com/amin20/GBM_WSSM.

Development and Validation of a Convolutional Neural Network for Automated Detection of Scaphoid Fractures on Conventional Radiographs.

To compare the performance of a convolutional neural network (CNN) to that of 11 radiologists in detecting scaphoid bone fractures on conventional radiographs of the hand, wrist, and scaphoid. At two hospitals (hospitals A and B), three datasets consisting of conventional hand, wrist, and scaphoid radiographs were retrospectively retrieved: a dataset of 1039 radiographs (775 patients [mean age, 48 years ± 23 {standard deviation}; 505 female patients], period: 2017-2019, hospitals A and B) for developing a scaphoid segmentation CNN, a dataset of 3000 radiographs (1846 patients [mean age, 42 years ± 22; 937 female patients], period: 2003-2019, hospital B) for developing a scaphoid fracture detection CNN, and a dataset of 190 radiographs (190 patients [mean age, 43 years ± 20; 77 female patients], period: 2011-2020, hospital A) for testing the complete fracture detection system. Both CNNs were applied consecutively: The segmentation CNN localized the scaphoid and then passed the relevant region to the detection CNN for fracture detection. In an observer study, the performance of the system was compared with that of 11 radiologists. Evaluation metrics included the Dice similarity coefficient (DSC), Hausdorff distance (HD), sensitivity, specificity, positive predictive value (PPV), and area under the receiver operating characteristic curve (AUC). The segmentation CNN achieved a DSC of 97.4% ± 1.4 with an HD of 1.31 mm ± 1.03. The detection CNN had sensitivity of 78% (95% CI: 70, 86), specificity of 84% (95% CI: 77, 92), PPV of 83% (95% CI: 77, 90), and AUC of 0.87 (95% CI: 0.81, 0.91). There was no difference between the AUC of the CNN and that of the radiologists (0.87 [95% CI: 0.81, 0.91] vs 0.83 [radiologist range: 0.79-0.85]; P = .09). The developed CNN achieved radiologist-level performance in detecting scaphoid bone fractures on conventional radiographs of the hand, wrist, and scaphoid.Keywords: Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning Algorithms, Feature Detection-Vision-Application Domain, Computer-Aided DiagnosisSee also the commentary by Li and Torriani in this issue.Supplemental material is available for this article.©RSNA, 2021.

Automated fiducial marker detection and localization in volumetric computed tomography images: a three-step hybrid approach with deep learning.

.Purpose: Automating fiducial detection and localization in the patient’s pre-operative images can lead to better registration accuracy, reduced human errors, and shorter intervention time. Most current approaches are optimized for a single marker type, mainly spherical adhesive markers. A fully automated algorithm is proposed and evaluated for screw and spherical titanium fiducials, typically used in high-accurate frameless surgical navigation.Approach: The algorithm builds on previous approaches with morphological functions and pose estimation algorithms. A 3D convolutional neural network (CNN) is proposed for the fiducial classification task and evaluated for both traditional closed-set and emerging open-set classifiers. A proposed digital ground-truth experiment, with cone-beam computed tomography (CBCT) imaging software, is performed to determine the localization accuracy of the algorithm. The localized fiducial positions in the CBCT images by the presented algorithm were compared to the actual known positions in the virtual phantom models. The difference represents the fiducial localization error (FLE).Results: A total of 241 screws, 151 spherical fiducials, and 1550 other structures are identified with the best true positive rate 95.9% for screw and 99.3% for spherical fiducials at 8.7% and 3.4% false positive rate, respectively. The best achieved FLE mean and its standard deviation for a screw and spherical marker are 58 (14) and , respectively.Conclusions: Accurate marker detection and localization were achieved, with spherical fiducials being superior to screws. Large marker volume and smaller voxel size yield significantly smaller FLEs. Attenuating noise by mesh smoothing has a minor effect on FLE. Future work will focus on expanding the CNN for image segmentation.

574 A deep learning classifier for the analysis of tiger tail banding in trichothiodystrophy

Trichothiodystrophy (TTD) and xeroderma pigmentosum (XP) are both rare, autosomal recessive disorders of DNA repair/ transcription genes. Despite having (different) mutations in the same genes, they have markedly different phenotypes. TTD patients have multisystem developmental abnormalities involving fetal growth, and brain, eyes, bones and immune system with normal cancer risk. XP patients have normal development and a greatly increased risk of sunlight induced skin cancer. The hallmark of TTD hair is an alternating dark and light “tiger tail” banding pattern under polarized light microscopy along with defects of the hair shaft and reduced content of sulfur containing amino acids. XP patients have normal hair. Some XP/TTD complex patients may have features of both disorders making the diagnosis and evaluation of cancer risk difficult. Here we present a U-Net convolutional neural network for segmentation and classification of tiger tail hair and normal hair images from polarized light microscopy. The model was trained with 35 TTD and 35 normal hair images. The dataset was augmented using standard crops, flips, and rotations. Shears were excluded to preserve the banded features. In future studies, a third class consisting of hair images from XP/TTD complex patients will be introduced to the U-Net to evaluate the differences in banding patterns. Use of artificial intelligence deep learning may be able to assist in diagnosis of patients with different types of hair abnormalities.

Multi-level graph convolutional recurrent neural network for semantic image segmentation

With the advent of the Internet of Things (IoT) era, many devices have surfaced that capture and generate various visual data. To recognize and extract a meaningful pattern from these visual data, powerful methods are required for different IoT applications. Fortunately, deep convolutional neural networks (CNNs) significantly improve the performance of almost all tasks in computer vision, including semantic image segmentation. However, the feature extraction of CNNs may cause the loss of contextual and spatial information. Moreover, the standard convolutional and pooling layers adopted by most CNN architectures lead to a fixed receptive field, which makes it challenging to deal with multi-scale objects in the image. To remedy these issues of CNNs for semantic image segmentation, this paper proposes a multi-level graph convolutional recurrent neural network (MGCRNN) to combine CNNs and graph neural networks (GNNs) for fusing multi-level features. By applying graph convolutional recurrent neural network (GCRNN), the proposed model acquires a global view of the image and aggregates multi-level contextual and structural information. The experiments verify the ability of GCRNN to obtain a flexible receptive field and learn structure features without losing spatial information. Results of these experiments conducted on the Pascal VOC 2012 and Cityscapes datasets show that the proposed model outperforms baseline approaches and can be competitive with state-of-the-art methods

Precise whole liver automatic segmentation and quantification of PDFF and R2* on MR images.

To automate the segmentation of whole liver parenchyma on multi-echo chemical shift encoded (MECSE) MR examinations using convolutional neural networks (CNNs) to seamlessly quantify precise organ-related imaging biomarkers such as the fat fraction and iron load. A retrospective multicenter collection of 183 MECSE liver MR examinations was conducted. An encoder-decoder CNN was trained (107 studies) following a 5-fold cross-validation strategy to improve the model performance and ensure lack of overfitting. Proton density fat fraction (PDFF) and R2* were quantified on both manual and CNN segmentation masks. Different metrics were used to evaluate the CNN performance over both unseen internal (46 studies) and external (29 studies) validation datasets to analyze reproducibility. The internal test showed excellent results for the automatic segmentation with a dice coefficient (DC) of 0.93 ± 0.03 and high correlation between the quantification done with the predicted mask and the manual segmentation (rPDFF = 1 and rR2* = 1; p values < 0.001). The external validation was also excellent with a different vendor but the same magnetic field strength, proving the generalization of the model to other manufacturers with DC of 0.94 ± 0.02. Results were lower for the 1.5-T MR same vendor scanner with DC of 0.87 ± 0.06. Both external validations showed high correlation in the quantification (rPDFF = 1 and rR2* = 1; p values < 0.001). In both internal and external validation datasets, the relative error for the PDFF and R2* quantification was below 4% and 1% respectively. Liver parenchyma can be accurately segmented with CNN in a vendor-neutral virtual approach, allowing to obtain reproducible automatic whole organ virtual biopsies. • Whole liver parenchyma can be automatically segmented using convolutional neural networks. • Deep learning allows the creation of automatic pipelines for the precise quantification of liver-related imaging biomarkers such as PDFF and R2*. • MR "virtual biopsy" can become a fast and automatic procedure for the assessment of chronic diffuse liver diseases in clinical practice.

Automated renal segmentation in healthy and chronic kidney disease subjects using a convolutional neural network.

Total kidney volume (TKV) is an important measure in renal disease detection and monitoring. We developed a fully automated method to segment the kidneys from T2 -weighted MRI to calculate TKV of healthy control (HC) and chronic kidney disease (CKD) patients. This automated method uses machine learning, specifically a 2D convolutional neural network (CNN), to accurately segment the left and right kidneys from T2 -weighted MRI data. The data set consisted of 30 HC subjects and 30 CKD patients. The model was trained on 50 manually defined HC and CKD kidney segmentations. The model was subsequently evaluated on 50 test data sets, comprising data from 5 HCs and 5 CKD patients each scanned 5 times in a scan session to enable comparison of the precision of the CNN and manual segmentation of kidneys. The unseen test data processed by the 2D CNN had a mean Dice score of 0.93 ± 0.01. The difference between manual and automatically computed TKV was 1.2 ± 16.2 mL with a mean surface distance of 0.65 ± 0.21 mm. The variance in TKV measurements from repeat acquisitions on the same subject was significantly lower using the automated method compared to manual segmentation of the kidneys. The 2D CNN method provides fully automated segmentation of the left and right kidney and calculation of TKV in <10 s on a standard office computer, allowing high data throughput and is a freely available executable.

Automatic aerospace weld inspection using unsupervised local deep feature learning

Automatic industrial inspection is critical to modern manufacturing enterprises. Due to their low cost and real-time processing speed, vision-based inspection systems are often used for this task. Deep Convolutional Neural Networks (CNNs) have been extensively applied to many computer vision tasks but require large numbers of annotated examples to train. Obtaining such annotations is expensive and time-consuming. In this study we describe a method which aims to make the best use of unannotated image data, which can often be collected easily. We propose a novel unsupervised local deep feature learning method based on image segmentation to build a network which can extract useful features from an image. The training algorithm alternates between (1) obtaining pseudo-labels by clustering the features extracted using a segmentation CNN and (2) training the CNN for feature learning using these pseudo-labels. To our knowledge, unsupervised local deep feature learning has not been addressed based on image segmentation in this way before. We demonstrate the approach on two aerospace weld inspection tasks. Our results show that the proposed unsupervised method performs almost as well as a CNN with the same architecture trained in a supervised manner.

Tumor attention networks: Better feature selection, better tumor segmentation

Compared with the traditional analysis of computed tomography scans, automatic liver tumor segmentation can supply precise tumor volumes and reduce the inter-observer variability in estimating the tumor size and the tumor burden, which could further assist physicians to make better therapeutic choices for hepatic diseases and monitoring treatment. Among current mainstream segmentation approaches, multi-layer and multi-kernel convolutional neural networks (CNNs) have attracted much attention in diverse biomedical/medical image segmentation tasks with remarkable performance. However, an arbitrary stacking of feature maps makes CNNs quite inconsistent in imitating the cognition and the visual attention of human beings for a specific visual task. To mitigate the lack of a reasonable feature selection mechanism in CNNs, we exploit a novel and effective network architecture, called Tumor Attention Networks (TA-Net), for mining adaptive features by embedding Tumor Attention layers with multi-functional modules to assist the liver tumor segmentation task. In particular, each tumor attention layer can adaptively highlight valuable tumor features and suppress unrelated ones among feature maps from 3D and 2D perspectives. Moreover, an analysis of visualization results illustrates the effectiveness of our tumor attention modules and the interpretability of CNNs for liver tumor segmentation. Furthermore, we explore different arrangements of skip connections in information fusion. A deep ablation study is also conducted to illustrate the effects of different attention strategies for hepatic tumors. The results of extensive experiments demonstrate that the proposed TA-Net increases the liver tumor segmentation performance with a lower computation cost and a small parameter overhead over the state-of-the-art methods, under various evaluation metrics on clinical benchmark data. In addition, two additional medical image datasets are used to evaluate generalization capability of TA-Net, including the comparison with general semantic segmentation methods and a non-tumor segmentation task. All the program codes have been released at https://github.com/shuchao1212/TA-Net.

Deep neural network-based detection and segmentation of intracranial aneurysms on 3D rotational DSA.

Accurate diagnosis and measurement of intracranial aneurysms are challenging. This study aimed to develop a 3D convolutional neural network (CNN) model to detect and segment intracranial aneurysms (IA) on 3D rotational DSA (3D-RA) images. 3D-RA images were collected and annotated by 5 neuroradiologists. The annotated images were then divided into three datasets: training, validation, and test. A 3D Dense-UNet-like CNN (3D-Dense-UNet) segmentation algorithm was constructed and trained using the training dataset. Diagnostic performance to detect aneurysms and segmentation accuracy was assessed for the final model on the test dataset using the free-response receiver operating characteristic (FROC). Finally, the CNN-inferred maximum diameter was compared against expert measurements by Pearson's correlation and Bland-Altman limits of agreement (LOA). A total of 451 patients with 3D-RA images were split into n = 347/41/63 training/validation/test datasets, respectively. For aneurysm detection, observed FROC analysis showed that the model managed to attain a sensitivity of 0.710 at 0.159 false positives (FP)/case, and 0.986 at 1.49 FP/case. The proposed method had good agreement with reference manual aneurysmal maximum diameter measurements (8.3 ± 4.3 mm vs. 7.8 ± 4.8 mm), with a correlation coefficient r = 0.77, small bias of 0.24 mm, and LOA of -6.2 to 5.71 mm. 37.0% and 77% of diameter measurements were within ±1 mm and ±2.5 mm of expert measurements. A 3D-Dense-UNet model can detect and segment aneurysms with relatively high accuracy using 3D-RA images. The automatically measured maximum diameter has potential clinical application value.

A comparison of methods for fully automatic segmentation of tumors and involved nodes in PET/CT of head and neck cancers

Target volume delineation is a vital but time-consuming and challenging part of radiotherapy, where the goal is to deliver sufficient dose to the target while reducing risks of side effects. For head and neck cancer (HNC) this is complicated by the complex anatomy of the head and neck region and the proximity of target volumes to organs at risk. The purpose of this study was to compare and evaluate conventional PET thresholding methods, six classical machine learning algorithms and a 2D U-Net convolutional neural network (CNN) for automatic gross tumor volume (GTV) segmentation of HNC in PET/CT images. For the latter two approaches the impact of single versus multimodality input on segmentation quality was also assessed. 197 patients were included in the study. The cohort was split into training and test sets (157 and 40 patients, respectively). Five-fold cross-validation was used on the training set for model comparison and selection. Manual GTV delineations represented the ground truth. Tresholding, classical machine learning and CNN segmentation models were ranked separately according to the cross-validation Sørensen–Dice similarity coefficient (Dice). PET thresholding gave a maximum mean Dice of 0.62, whereas classical machine learning resulted in maximum mean Dice scores of 0.24 (CT) and 0.66 (PET; PET/CT). CNN models obtained maximum mean Dice scores of 0.66 (CT), 0.68 (PET) and 0.74 (PET/CT). The difference in cross-validation Dice between multimodality PET/CT and single modality CNN models was significant (p ≤ 0.0001). The top-ranked PET/CT-based CNN model outperformed the best-performing thresholding and classical machine learning models, giving significantly better segmentations in terms of cross-validation and test set Dice, true positive rate, positive predictive value and surface distance-based metrics (p ≤ 0.0001). Thus, deep learning based on multimodality PET/CT input resulted in superior target coverage and less inclusion of surrounding normal tissue.

DeepMIB: User-friendly and open-source software for training of deep learning network for biological image segmentation.

We present DeepMIB, a new software package that is capable of training convolutional neural networks for segmentation of multidimensional microscopy datasets on any workstation. We demonstrate its successful application for segmentation of 2D and 3D electron and multicolor light microscopy datasets with isotropic and anisotropic voxels. We distribute DeepMIB as both an open-source multi-platform Matlab code and as compiled standalone application for Windows, MacOS and Linux. It comes in a single package that is simple to install and use as it does not require knowledge of programming. DeepMIB is suitable for everyone interested of bringing a power of deep learning into own image segmentation workflows.

Abstract PO-004: A deep convolutional neural network for segmentation of whole-slide pathology images in glioblastoma

Abstract Glioblastoma is the most aggressive type of brain cancer with high-levels of intra- and inter-tumour heterogeneity that contribute to its rapid growth and invasion within the brain. Here, we have used a deep convolutional neural network (DCNN) as a semantic segmentation model to segment seven different tumour regions including leading-edge (LE), infiltrating tumour (IT), cellular tumour (CT), cellular tumour microvascular proliferation (CTmvp), cellular tumour pseudopalisading region around necrosis (CTpan), cellular tumour perinecrotic zones (CTpnz) and cellular tumour necrosis (CTne) in digitised glioblastoma histopathological slides from The Cancer Genome Atlas (TCGA). Analysis of segmentation results from tumour images together with matched RNA expression data identified genetic signatures that are specific to these different tumour regions. We found that spatially resolved gene signatures were strongly correlated with survival in patients with defined genetic mutations. Moreover, in silico cell ontology analysis and single-cell RNA sequencing data from resected glioblastoma tissue samples, showed that these tumour regions had different gene signatures, suggesting they are driven by different cell types in the tumour microenvironment. This points to a key role for interactions between microglia/pericytes/monocytes and tumour cells that occur in the IT and CTmvp regions, which may contribute to poor patient survival. Overall, this work identifies key histopathological features that are indicative of patient survival and detected spatially associated genetic signatures that mediate tumour-stroma interactions that should be investigated as new targets in glioblastoma. Citation Format: Amin Zadeh Shirazi, Mark D. McDonnell, Eric Fornaciari, Narjes Sadat Bagherian, Kaitlin G. Scheer, Michael S. Samuel, Mahdi Yaghoobi, Rebecca J. Ormsby, Santosh Poonnoose, Damon Tumes, Guillermo A. Gomez. A deep convolutional neural network for segmentation of whole-slide pathology images in glioblastoma [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-004.

Learning Tubule-Sensitive CNNs for Pulmonary Airway and Artery-Vein Segmentation in CT.

Training convolutional neural networks (CNNs) for segmentation of pulmonary airway, artery, and vein is challenging due to sparse supervisory signals caused by the severe class imbalance between tubular targets and background. We present a CNNs-based method for accurate airway and artery-vein segmentation in non-contrast computed tomography. It enjoys superior sensitivity to tenuous peripheral bronchioles, arterioles, and venules. The method first uses a feature recalibration module to make the best use of features learned from the neural networks. Spatial information of features is properly integrated to retain relative priority of activated regions, which benefits the subsequent channel-wise recalibration. Then, attention distillation module is introduced to reinforce representation learning of tubular objects. Fine-grained details in high-resolution attention maps are passing down from one layer to its previous layer recursively to enrich context. Anatomy prior of lung context map and distance transform map is designed and incorporated for better artery-vein differentiation capacity. Extensive experiments demonstrated considerable performance gains brought by these components. Compared with state-of-the-art methods, our method extracted much more branches while maintaining competitive overall segmentation performance. Codes and models are available at http://www.pami.sjtu.edu.cn/News/56.