Deep learning ensembles for detecting brain metastases in longitudinal multi-modal MRI studies.

Metastatic brain cancer is a condition characterized by the migration of cancer cells to the brain from extracranial sites. Notably, metastatic brain tumors surpass primary brain tumors in prevalence by a significant factor, they exhibit an aggressive growth potential and have the capacity to spread across diverse cerebral locations simultaneously. Magnetic resonance imaging (MRI) scans of individuals afflicted with metastatic brain tumors unveil a wide spectrum of characteristics. These lesions vary in size and quantity, spanning from tiny nodules to substantial masses captured within MRI. Patients may present with a limited number of lesions or an extensive burden of hundreds of them. Moreover, longitudinal studies may depict surgical resection cavities, as well as areas of necrosis or edema. Thus, the manual analysis of such MRI scans is difficult, user-dependent and cost-inefficient, and - importantly - it lacks reproducibility. We address these challenges and propose a pipeline for detecting and analyzing brain metastases in longitudinal studies, which benefits from an ensemble of various deep learning architectures originally designed for different downstream tasks (detection and segmentation). The experiments, performed over 275 multi-modal MRI scans of 87 patients acquired in 53 sites, coupled with rigorously validated manual annotations, revealed that our pipeline, built upon open-source tools to ensure its reproducibility, offers high-quality detection, and allows for precisely tracking the disease progression. To objectively quantify the generalizability of models, we introduce a new data stratification approach that accommodates the heterogeneity of the dataset and is used to elaborate training-test splits in a data-robust manner, alongside a new set of quality metrics to objectively assess algorithms. Our system provides a fully automatic and quantitative approach that may support physicians in a laborious process of disease progression tracking and evaluation of treatment efficacy.

Machine learning identifies metabolic dysfunction associated steatotic liver disease in patients with diabetes mellitus.

BACKGROUND & AIMS: The presence of metabolic dysfunction associated steatotic liver disease (MASLD) in patients with diabetes mellitus (DM) is associated with a high risk of cardiovascular disease, but is often under-diagnosed. The objective is to develop machine learning (ML) models for risk assessment of MASLD occurrence in patients with DM. METHODS: Feature selection determined the discriminative parameters, utilized to classify DM patients as those with and without MASLD. The multiple logistic regression (MLR) model's performance was quantified by sensitivity, specificity, percentage of correctly classified patients, and receiver operating characteristic (ROC) curve analysis. Decision curve analysis (DCA) assessed the model's net benefit for alternative treatments. RESULTS: We studied 2000 patients with DM (mean age 58.85±17.37 years; 48% women). Eight parameters: age, body mass index, type of DM, alanine aminotransferase, aspartate aminotransferase, platelet count, hyperuricaemia, and treatment with metformin were identified as discriminative. The experiments for 1735 patients show that 744/991 (75.08%) and 586/744 (78.76%) patients with/without MASLD were correctly identified (sensitivity/specificity: 0.75/0.79). The area under ROC (AUC) was 0.84 (95%CI: 0.82-0.86), while DCA showed a higher model's clinical utility, ranging from 30-84% threshold probability. Results for 265 test patients confirm the model's generalizability (sensitivity/specificity: 0.80/0.74, AUC: 0.81 [95%CI: 0.76-0.87]), whereas unsupervised clustering identified high-risk patients. CONCLUSIONS: A ML approach demonstrated high performance in identifying MASLD in patients with DM. This approach may facilitate better risk stratification and cardiovascular risk prevention strategies for high-risk patients with DM at risk of MASLD.

Machine learning identification of risk factors for heart failure in patients with diabetes mellitus with metabolic dysfunction associated steatotic liver disease (MASLD): the Silesia Diabetes-Heart Project.

BACKGROUND: Diabetes mellitus (DM), heart failure (HF) and metabolic dysfunction associated steatotic liver disease (MASLD) are overlapping diseases of increasing prevalence. Because there are still high numbers of patients with HF who are undiagnosed and untreated, there is a need for improving efforts to better identify HF in patients with DM with or without MASLD. This study aims to develop machine learning (ML) models for assessing the risk of the HF occurrence in patients with DM with and without MASLD. RESEARCH DESIGN AND METHODS: In the Silesia Diabetes-Heart Project (NCT05626413), patients with DM with and without MASLD were analyzed to identify the most important HF risk factors with the use of a ML approach. The multiple logistic regression (MLR) classifier exploiting the most discriminative patient's parameters selected by the χ2 test following the Monte Carlo strategy was implemented. The classification capabilities of the ML models were quantified using sensitivity, specificity, and the percentage of correctly classified (CC) high- and low-risk patients. RESULTS: We studied 2000 patients with DM (mean age 58.85 ± SD 17.37 years; 48% women). In the feature selection process, we identified 5 parameters: age, type of DM, atrial fibrillation (AF), hyperuricemia and estimated glomerular filtration rate (eGFR). In the case of MASLD( +) patients, the same criterion was met by 3 features: AF, hyperuricemia and eGFR, and for MASLD(-) patients, by 2 features: age and eGFR. Amongst all patients, sensitivity and specificity were 0.81 and 0.70, respectively, with the area under the receiver operating curve (AUC) of 0.84 (95% CI 0.82-0.86). CONCLUSION: A ML approach demonstrated high performance in identifying HF in patients with DM independently of their MASLD status, as well as both in patients with and without MASLD based on easy-to-obtain patient parameters.

Clinical Trial: Probiotics in Metformin Intolerant Patients with Type 2 Diabetes (ProGasMet).

BACKGROUND: For decades, metformin has been the drug of first choice in the management of type 2 diabetes. However, approximately 2-13% of patients do not tolerate metformin due to gastrointestinal (GI) side effects. Since metformin influences the gut microbiota, we hypothesized that a multi-strain probiotics supplementation would mitigate the gastrointestinal symptoms associated with metformin usage. METHODS AND ANALYSIS: This randomized, double-blind, placebo-controlled, single-center, cross-over trial (ProGasMet study) assessed the efficacy of a multi-strain probiotic in 37 patients with metformin intolerance. Patients were randomly allocated (1:1) to receive probiotic (PRO-PLA) or placebo (PLA-PRO) at baseline and, after 12 weeks (period 1), they crossed-over to the other treatment arm (period 2). The primary outcome was the reduction of GI adverse events of metformin. RESULTS: 37 out of 82 eligible patients were enrolled in the final analysis of whom 35 completed the 32 weeks study period and 2 patients resigned at visit 5. Regardless of the treatment arm allocation, while on probiotic supplementation, there was a significant reduction of incidence (for the probiotic period in PRO-PLA/PLA-PRO: P = 0.017/P = 0.054), quantity and severity of nausea (P = 0.016/P = 0.024), frequency (P = 0.009/P = 0.015) and severity (P = 0.019/P = 0.005) of abdominal bloating/pain as well as significant improvement in self-assessed tolerability of metformin (P < 0.01/P = 0.005). Moreover, there was significant reduction of incidence of diarrhea while on probiotic supplementation in PRO-PLA treatment arm (P = 0.036). CONCLUSION: A multi-strain probiotic diminishes the incidence of gastrointestinal adverse effects in patients with type 2 diabetes and metformin intolerance.

A Real-World Benchmark for Sentinel-2 Multi-Image Super-Resolution.

Insufficient image spatial resolution is a serious limitation in many practical scenarios, especially when acquiring images at a finer scale is infeasible or brings higher costs. This is inherent to remote sensing, including Sentinel-2 satellite images that are available free of charge at a high revisit frequency, but whose spatial resolution is limited to 10m ground sampling distance. The resolution can be increased with super-resolution algorithms, in particular when performed from multiple images captured at subsequent revisits of a satellite, taking advantage of information fusion that leads to enhanced reconstruction accuracy. One of the obstacles in multi-image super-resolution consists in the scarcity of real-world benchmarks-commonly, simulated data are exploited which do not fully reflect the operating conditions. In this paper, we introduce a new benchmark (named MuS2) for super-resolving multiple Sentinel-2 images, with WorldView-2 imagery used as the high-resolution reference. Within MuS2, we publish the first end-to-end evaluation procedure for this problem which we expect to help the researchers in advancing the state of the art in multi-image super-resolution.

Machine learning profiles of cardiovascular risk in patients with diabetes mellitus: the Silesia Diabetes-Heart Project.

AIMS: As cardiovascular disease (CVD) is a leading cause of death for patients with diabetes mellitus (DM), we aimed to find important factors that predict cardiovascular (CV) risk using a machine learning (ML) approach. METHODS AND RESULTS: We performed a single center, observational study in a cohort of 238 DM patients (mean age ± SD 52.15 ± 17.27 years, 54% female) as a part of the Silesia Diabetes-Heart Project. Having gathered patients' medical history, demographic data, laboratory test results, results from the Michigan Neuropathy Screening Instrument (assessing diabetic peripheral neuropathy) and Ewing's battery examination (determining the presence of cardiovascular autonomic neuropathy), we managed use a ML approach to predict the occurrence of overt CVD on the basis of five most discriminative predictors with the area under the receiver operating characteristic curve of 0.86 (95% CI 0.80-0.91). Those features included the presence of past or current foot ulceration, age, the treatment with beta-blocker (BB) and angiotensin converting enzyme inhibitor (ACEi). On the basis of the aforementioned parameters, unsupervised clustering identified different CV risk groups. The highest CV risk was determined for the eldest patients treated in large extent with ACEi but not BB and having current foot ulceration, and for slightly younger individuals treated extensively with both above-mentioned drugs, with relatively small percentage of diabetic ulceration. CONCLUSIONS: Using a ML approach in a prospective cohort of patients with DM, we identified important factors that predicted CV risk. If a patient was treated with ACEi or BB, is older and has/had a foot ulcer, this strongly predicts that he/she is at high risk of having overt CVD.

Early detection of Solanum lycopersicum diseases from temporally-aggregated hyperspectral measurements using machine learning.

Some plant diseases can significantly reduce harvest, but their early detection in cultivation may prevent those consequential losses. Conventional methods of diagnosing plant diseases are based on visual observation of crops, but the symptoms of various diseases may be similar. It increases the difficulty of this task even for an experienced farmer and requires detailed examination based on invasive methods conducted in laboratory settings by qualified personnel. Therefore, modern agronomy requires the development of non-destructive crop diagnosis methods to accelerate the process of detecting plant infections with various pathogens. This research pathway is followed in this paper, and an approach for classifying selected Solanum lycopersicum diseases (anthracnose, bacterial speck, early blight, late blight and septoria leaf) from hyperspectral data captured on consecutive days post inoculation (DPI) is presented. The objective of that approach was to develop a technique for detecting infection in less than seven days after inoculation. The dataset used in this study included hyperspectral measurements of plants of two cultivars of S. lycopersicum: Benito and Polfast, which were infected with five different pathogens. Hyperspectral reflectance measurements were performed using a high-spectral-resolution field spectroradiometer (350-2500 nm range) and they were acquired for 63 days after inoculation, with particular emphasis put on the first 17 day-by-day measurements. Due to a significant data imbalance and low representation of measurements on some days, the collective datasets were elaborated by combining measurements from several days. The experimental results showed that machine learning techniques can offer accurate classification, and they indicated the practical utility of our approaches.

Relationship of vitamin D deficiency with cardiovascular disease and glycemic control in patients with type 2 diabetes mellitus: the Silesia Diabetes-Heart Project.

INTRODUCTION: Vitamin D (VD) has a pleiotropic effect on many health­related aspects, yet the results of studies regarding vitamin D deficiency (VDD) and both glycemic control and cardiovascular disease (CVD) are conflicting. OBJECTIVE: The aim of this work was to determine the prevalence of VDD and its associations with CVD and glycemic control among patients with type 2 diabetes mellitus (T2DM). PATIENTS AND METHODS: This was an observational study in T2DM patients recruited at the diabetology clinic in Zabrze, Poland (April-September 2019 and April-September 2020). The presence of CVD was determined based on medical records. Blood biochemical parameters, densitometry, and carotid artery ultrasound examination were performed. Control of diabetes was assessed based on glycated hemoglobin A1c (HbA1c) levels. A serum VD level below 20 ng/ml was considered as VDD. RESULTS: The prevalence of VDD in 197 patients was 36%. CVD was evident in 27% of the patients with VDD and in 33% of the patients with VD within the normal range (vitamin D sufficiency [VDS]) (P = 0.34). The difference between the groups regarding diabetes control was insignificant (P = 0.05), as for the VDD patients the median value (interquartile range) of HbA1c was 7.5% (6.93%-7.9%), and for VDS patients it was 7.5% (6.56%-7.5%). The VDD patients were more often treated with sodium­glucose cotransporter­2 inhibitors (SGLT­2is) (44% vs 25%; P = 0.01). CONCLUSIONS: About one­third of the patients showed VDD. The VDD and VDS groups did not differ in terms of CVD occurrence and the difference in glycemic control was insignificant. The patients with VDD were more often treated with SGLT­2is, which requires further investigation.

Machine Learning Predicts Cardiovascular Events in Patients With Diabetes: The Silesia Diabetes-Heart Project.

We aimed to develop a machine learning (ML) model for predicting cardiovascular (CV) events in patients with diabetes (DM). This was a prospective, observational study where clinical data of patients with diabetes hospitalized in the diabetology center in Poland (years 2015-2020) were analyzed using ML. The occurrence of new CV events following discharge was collected in the follow-up time for up to 5 years and 9 months. An end-to-end ML technique which exploits the neighborhood component analysis for elaborating discriminative predictors, followed by a hybrid sampling/boosting classification algorithm, multiple logistic regression (MLR), or unsupervised hierarchical clustering was proposed. In 1735 patients with diabetes (53% female), there were 150 (8.65%) ones with a new CV event in the follow-up. Twelve most discriminative patients' parameters included coronary artery disease, heart failure, peripheral artery disease, stroke, diabetic foot disease, chronic kidney disease, eosinophil count, serum potassium level, and being treated with clopidogrel, heparin, proton pump inhibitor, and loop diuretic. Utilizing those variables resulted in the area under the receiver operating characteristic curve (AUC) ranging from 0.62 (95% Confidence Interval [CI] 0.56-0.68, P < 0.01) to 0.72 (95% CI 0.66-0.77, P < 0.01) across 5 nonoverlapping test folds, whereas MLR correctly determined 111/150 (74.00%) high-risk patients, and 989/1585 (62.40%) low-risk patients, resulting in 1100/1735 (63.40%) correctly classified patients (AUC: 0.72, 95% CI 0.66-0.77). ML algorithms can identify patients with diabetes at a high risk of new CV events based on a small number of interpretable and easy-to-obtain patients' parameters.

Deep learning automates bidimensional and volumetric tumor burden measurement from MRI in pre- and post-operative glioblastoma patients.

Tumor burden assessment by magnetic resonance imaging (MRI) is central to the evaluation of treatment response for glioblastoma. This assessment is, however, complex to perform and associated with high variability due to the high heterogeneity and complexity of the disease. In this work, we tackle this issue and propose a deep learning pipeline for the fully automated end-to-end analysis of glioblastoma patients. Our approach simultaneously identifies tumor sub-regions, including the enhancing tumor, peritumoral edema and surgical cavity in the first step, and then calculates the volumetric and bidimensional measurements that follow the current Response Assessment in Neuro-Oncology (RANO) criteria. Also, we introduce a rigorous manual annotation process which was followed to delineate the tumor sub-regions by the human experts, and to capture their segmentation confidences that are later used while training deep learning models. The results of our extensive experimental study performed over 760 pre-operative and 504 post-operative adult patients with glioma obtained from the public database (acquired at 19 sites in years 2021-2020) and from a clinical treatment trial (47 and 69 sites for pre-/post-operative patients, 2009-2011) and backed up with thorough quantitative, qualitative and statistical analysis revealed that our pipeline performs accurate segmentation of pre- and post-operative MRIs in a fraction of the manual delineation time (up to 20 times faster than humans). Volumetric measurements were in strong agreement with experts with the Intraclass Correlation Coefficient (ICC): 0.959, 0.703, 0.960 for ET, ED, and cavity. Similarly, automated RANO compared favorably with experienced readers (ICC: 0.681 and 0.866) producing consistent and accurate results. Additionally, we showed that RANO measurements are not always sufficient to quantify tumor burden. The high performance of the automated tumor burden measurement highlights the potential of the tool for considerably improving and simplifying radiological evaluation of glioblastoma in clinical trials and clinical practice.

Detecting liver cirrhosis in computed tomography scans using clinically-inspired and radiomic features.

Hepatic cirrhosis is an increasing cause of mortality in developed countries-it is the pathological sequela of chronic liver diseases, and the final liver fibrosis stage. Since cirrhosis evolves from the asymptomatic phase, it is of paramount importance to detect it as quickly as possible, because entering the symptomatic phase commonly leads to hospitalization and can be fatal. Understanding the state of the liver based on the abdominal computed tomography (CT) scans is tedious, user-dependent and lacks reproducibility. We tackle these issues and propose an end-to-end and reproducible approach for detecting cirrhosis from CT. It benefits from the introduced clinically-inspired features that reflect the patient's characteristics which are often investigated by experienced radiologists during the screening process. Such features are coupled with the radiomic ones extracted from the liver, and from the suggested region of interest which captures the liver's boundary. The rigorous experiments, performed over two heterogeneous clinical datasets (two cohorts of 241 and 32 patients) revealed that extracting radiomic features from the liver's rectified contour is pivotal to enhance the classification abilities of the supervised learners. Also, capturing clinically-inspired image features significantly improved the performance of such models, and the proposed features were consistently selected as the important ones. Finally, we showed that selecting the most discriminative features leads to the Pareto-optimal models with enhanced feature-level interpretability, as the number of features was dramatically reduced (280×) from thousands to tens.

Risk factors for cardiovascular disease in patients with metabolic-associated fatty liver disease: a machine learning approach.

BACKGROUND: Nonalcoholic fatty liver disease is associated with an increased cardiovascular disease (CVD) risk, although the exact mechanism(s) are less clear. Moreover, the relationship between newly redefined metabolic-associated fatty liver disease (MAFLD) and CVD risk has been poorly investigated. Data-driven machine learning (ML) techniques may be beneficial in discovering the most important risk factors for CVD in patients with MAFLD. METHODS: In this observational study, the patients with MAFLD underwent subclinical atherosclerosis assessment and blood biochemical analysis. Patients were split into two groups based on the presence of CVD (defined as at least one of the following: coronary artery disease; myocardial infarction; coronary bypass grafting; stroke; carotid stenosis; lower extremities artery stenosis). The ML techniques were utilized to construct a model which could identify individuals with the highest risk of CVD. We exploited the multiple logistic regression classifier operating on the most discriminative patient's parameters selected by univariate feature ranking or extracted using principal component analysis (PCA). Receiver operating characteristic (ROC) curves and area under the ROC curve (AUC) were calculated for the investigated classifiers, and the optimal cut-point values were extracted from the ROC curves using the Youden index, the closest to (0, 1) criteria and the Index of Union methods. RESULTS: In 191 patients with MAFLD (mean age: 58, SD: 12 years; 46% female), there were 47 (25%) patients who had the history of CVD. The most important clinical variables included hypercholesterolemia, the plaque scores, and duration of diabetes. The five, ten and fifteen most discriminative parameters extracted using univariate feature ranking and utilized to fit the ML models resulted in AUC of 0.84 (95% confidence interval [CI]: 0.77-0.90, p < 0.0001), 0.86 (95% CI 0.80-0.91, p < 0.0001) and 0.87 (95% CI 0.82-0.92, p < 0.0001), whereas the classifier fitted over 10 principal components extracted using PCA followed by the parallel analysis obtained AUC of 0.86 (95% CI 0.81-0.91, p < 0.0001). The best model operating on 5 most discriminative features correctly identified 114/144 (79.17%) low-risk and 40/47 (85.11%) high-risk patients. CONCLUSION: A ML approach demonstrated high performance in identifying MAFLD patients with prevalent CVD based on the easy-to-obtain patient parameters.

First Gradually, Then Suddenly: Understanding the Impact of Image Compression on Object Detection Using Deep Learning.

Video surveillance systems process high volumes of image data. To enable long-term retention of recorded images and because of the data transfer limitations in geographically distributed systems, lossy compression is commonly applied to images prior to processing, but this causes a deterioration in image quality due to the removal of potentially important image details. In this paper, we investigate the impact of image compression on the performance of object detection methods based on convolutional neural networks. We focus on Joint Photographic Expert Group (JPEG) compression and thoroughly analyze a range of the performance metrics. Our experimental study, performed over a widely used object detection benchmark, assessed the robustness of nine popular object-detection deep models against varying compression characteristics. We show that our methodology can allow practitioners to establish an acceptable compression level for specific use cases; hence, it can play a key role in applications that process and store very large image data.

Segmenting pediatric optic pathway gliomas from MRI using deep learning.

Optic pathway gliomas are low-grade neoplastic lesions that account for approximately 3-5% of brain tumors in children. Assessing tumor burden from magnetic resonance imaging (MRI) plays a central role in its efficient management, yet it is a challenging and human-dependent task due to the difficult and error-prone process of manual segmentation of such lesions, as they can easily manifest different location and appearance characteristics. In this paper, we tackle this issue and propose a fully-automatic and reproducible deep learning algorithm built upon the recent advances in the field which is capable of detecting and segmenting optical pathway gliomas from MRI. The proposed training strategies help us elaborate well-generalizing deep models even in the case of limited ground-truth MRIs presenting example optic pathway gliomas. The rigorous experimental study, performed over two clinical datasets of 22 and 51 multi-modal MRIs acquired for 22 and 51 patients with optical pathway gliomas, and a public dataset of 494 pre-surgery low-/high-grade glioma patients (corresponding to 494 multi-modal MRIs), and involving quantitative, qualitative and statistical analysis revealed that the suggested technique can not only effectively delineate optic pathway gliomas, but can also be applied for detecting other brain tumors. The experiments indicate high agreement between automatically calculated and ground-truth volumetric measurements of the tumors and very fast operation of the proposed approach, both of which can increase the clinical utility of the suggested software tool. Finally, our deep architectures have been made open-sourced to ensure full reproducibility of the method over other MRI data.

Recent Advances in Multi- and Hyperspectral Image Analysis.

Current advancements in sensor technology bring new possibilities in multi- and hyperspectral imaging. Real-life use cases which can benefit from such imagery span across various domains, including precision agriculture, chemistry, biology, medicine, land cover applications, management of natural resources, detecting natural disasters, and more. To extract value from such highly dimensional data capturing up to hundreds of spectral bands in the electromagnetic spectrum, researchers have been developing a range of image processing and machine learning analysis pipelines to process these kind of data as efficiently as possible. To this end, multi- or hyperspectral analysis has bloomed and has become an exciting research area which can enable the faster adoption of this technology in practice, also when such algorithms are deployed in hardware-constrained and extreme execution environments; e.g., on-board imaging satellites.

Dynamics of facial actions for assessing smile genuineness.

Applying computer vision techniques to distinguish between spontaneous and posed smiles is an active research topic of affective computing. Although there have been many works published addressing this problem and a couple of excellent benchmark databases created, the existing state-of-the-art approaches do not exploit the action units defined within the Facial Action Coding System that has become a standard in facial expression analysis. In this work, we explore the possibilities of extracting discriminative features directly from the dynamics of facial action units to differentiate between genuine and posed smiles. We report the results of our experimental study which shows that the proposed features offer competitive performance to those based on facial landmark analysis and on textural descriptors extracted from spatial-temporal blocks. We make these features publicly available for the UvA-NEMO and BBC databases, which will allow other researchers to further improve the classification scores, while preserving the interpretation capabilities attributed to the use of facial action units. Moreover, we have developed a new technique for identifying the smile phases, which is robust against the noise and allows for continuous analysis of facial videos.

Fully-automated deep learning-powered system for DCE-MRI analysis of brain tumors.

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays an important role in diagnosis and grading of brain tumors. Although manual DCE biomarker extraction algorithms boost the diagnostic yield of DCE-MRI by providing quantitative information on tumor prognosis and prediction, they are time-consuming and prone to human errors. In this paper, we propose a fully-automated, end-to-end system for DCE-MRI analysis of brain tumors. Our deep learning-powered technique does not require any user interaction, it yields reproducible results, and it is rigorously validated against benchmark and clinical data. Also, we introduce a cubic model of the vascular input function used for pharmacokinetic modeling which significantly decreases the fitting error when compared with the state of the art, alongside a real-time algorithm for determination of the vascular input region. An extensive experimental study, backed up with statistical tests, showed that our system delivers state-of-the-art results while requiring less than 3 min to process an entire input DCE-MRI study using a single GPU.

Segmenting brain tumors from FLAIR MRI using fully convolutional neural networks.

BACKGROUND AND OBJECTIVE: Magnetic resonance imaging (MRI) is an indispensable tool in diagnosing brain-tumor patients. Automated tumor segmentation is being widely researched to accelerate the MRI analysis and allow clinicians to precisely plan treatment-accurate delineation of brain tumors is a critical step in assessing their volume, shape, boundaries, and other characteristics. However, it is still a very challenging task due to inherent MR data characteristics and high variability, e.g., in tumor sizes or shapes. We present a new deep learning approach for accurate brain tumor segmentation which can be trained from small and heterogeneous datasets annotated by a human reader (providing high-quality ground-truth segmentation is very costly in practice). METHODS: In this paper, we present a new deep learning technique for segmenting brain tumors from fluid attenuation inversion recovery MRI. Our technique exploits fully convolutional neural networks, and it is equipped with a battery of augmentation techniques that make the algorithm robust against low data quality, and heterogeneity of small training sets. We train our models using only positive (tumorous) examples, due to the limited amount of available data. RESULTS: Our algorithm was tested on a set of stage II-IV brain-tumor patients (image data collected using MAGNETOM Prisma 3T, Siemens). Rigorous experiments, backed up with statistical tests, revealed that our approach outperforms the state-of-the-art approach (utilizing hand-crafted features) in terms of segmentation accuracy, offers very fast training and instant segmentation (analysis of an image takes less than a second). Building our deep model is 1.3 times faster compared with extracting features for extremely randomized trees, and this training time can be controlled. Finally, we showed that too aggressive data augmentation may lead to deteriorated performance of the model, especially in the fixed-budget training (with maximum numbers of training epochs). CONCLUSIONS: Our method yields the better performance when compared with the state of the art method which utilizes hand-crafted features. In addition, our deep network can be effectively applied to difficult (small, imbalanced, and heterogeneous) datasets, offers controllable training time, and infers in real-time.

Data Augmentation for Brain-Tumor Segmentation: A Review.

Data augmentation is a popular technique which helps improve generalization capabilities of deep neural networks, and can be perceived as implicit regularization. It plays a pivotal role in scenarios in which the amount of high-quality ground-truth data is limited, and acquiring new examples is costly and time-consuming. This is a very common problem in medical image analysis, especially tumor delineation. In this paper, we review the current advances in data-augmentation techniques applied to magnetic resonance images of brain tumors. To better understand the practical aspects of such algorithms, we investigate the papers submitted to the Multimodal Brain Tumor Segmentation Challenge (BraTS 2018 edition), as the BraTS dataset became a standard benchmark for validating existent and emerging brain-tumor detection and segmentation techniques. We verify which data augmentation approaches were exploited and what was their impact on the abilities of underlying supervised learners. Finally, we highlight the most promising research directions to follow in order to synthesize high-quality artificial brain-tumor examples which can boost the generalization abilities of deep models.