Adrenal lesion classification with abdomen caps and the effect of ROI size.

Accurate classification of adrenal lesions on magnetic resonance (MR) images are very important for diagnosis and treatment planning. The detection and classification of lesions in medical imaging heavily rely on several key factors, including the specialist's level of experience, work intensity, and fatigue of the clinician. These factors are critical determinants of the accuracy and effectiveness of the diagnostic process, which in turn has a direct impact on patient health outcomes. With the spread of artificial intelligence, the use of computer-aided diagnosis (CAD) systems in disease diagnosis has also increased. In this study, adrenal lesion classification was performed using deep learning on MR images. The data set used was obtained from the Department of Radiology, Faculty of Medicine, Selcuk University, and all adrenal lesions were identified and reviewed in consensus by two radiologists experienced with abdominal MR. Studies were carried out on two different data sets created by T1- and T2-weighted MR images. The data set consisted of 112 benign and 10 malignant lesions for each mode. Experiments were performed with regions of interest (ROIs) of different sizes to increase the working performance. Thus, the effect of the selected ROI size on the classification performance was assessed. In addition, instead of the convolutional neural network (CNN) models used in deep learning, a unique classification model structure called Abdomen Caps was proposed. When the data sets used in classification studies are manually separated for training, validation, and testing, different results are obtained with different data sets for each stage. To eliminate this imbalance, tenfold cross-validation was used in this study. The best results obtained were 0.982, 0.999, 0.969, 0.983, 0.998, and 0.964 for accuracy, precision, recall, F1-score, area under the curve (AUC) score, and kappa score, respectively.

An extensive study for binary characterisation of adrenal tumours.

On adrenal glands, benign tumours generally change the hormone equilibrium, and malign tumours usually tend to spread to the nearby tissues and to the organs of the immune system. These features can give a trace about the type of adrenal tumours; however, they cannot be observed all the time. Different tumour types can be confused in terms of having a similar shape, size and intensity features on scans. To support the evaluation process, biopsy process is applied that includes injury and complication risks. In this study, we handle the binary characterisation of adrenal tumours by using dynamic computed tomography images. Concerning this, the usage of one more imaging modalities and biopsy process is wanted to be excluded. The used dataset consists of 8 subtypes of adrenal tumours, and it seemed as the worst-case scenario in which all handicaps are available against tumour classification. Histogram, grey level co-occurrence matrix and wavelet-based features are investigated to reveal the most effective one on the identification of adrenal tumours. Binary classification is proposed utilising four-promising algorithms that have proven oneself on the task of binary-medical pattern classification. For this purpose, optimised neural networks are examined using six dataset inspired by the aforementioned features, and an efficient framework is offered before the use of a biopsy. Accuracy, sensitivity, specificity, and AUC are used to evaluate the performance of classifiers. Consequently, malign/benign characterisation is performed by proposed framework, with success rates of 80.7%, 75%, 82.22% and 78.61% for the metrics, respectively. Graphical abstract.

Adrenal tumor segmentation method for MR images.

BACKGROUND AND OBJECTIVE: Adrenal tumors, which occur on adrenal glands, are incidentally determined. The liver, spleen, spinal cord, and kidney surround the adrenal glands. Therefore, tumors on the adrenal glands can be adherent to other organs. This is a problem in adrenal tumor segmentation. In addition, low contrast, non-standardized shape and size, homogeneity, and heterogeneity of the tumors are considered as problems in segmentation. METHODS: This study proposes a computer-aided diagnosis (CAD) system to segment adrenal tumors by eliminating the above problems. The proposed hybrid method incorporates many image processing methods, which include active contour, adaptive thresholding, contrast limited adaptive histogram equalization (CLAHE), image erosion, and region growing. RESULTS: The performance of the proposed method was assessed on 113 Magnetic Resonance (MR) images using seven metrics: sensitivity, specificity, accuracy, precision, Dice Coefficient, Jaccard Rate, and structural similarity index (SSIM). The proposed method eliminates some of the discussed problems with success rates of 74.84%, 99.99%, 99.84%, 93.49%, 82.09%, 71.24%, 99.48% for the metrics, respectively. CONCLUSIONS: This study presents a new method for adrenal tumor segmentation, and avoids some of the problems preventing accurate segmentation, especially for cyst-based tumors.

A novel pipeline for adrenal tumour segmentation.

BACKGROUND AND OBJECTIVE: Adrenal tumours occur on adrenal glands surrounded by organs and osteoid. These tumours can be categorized as either functional, non-functional, malign, or benign. Depending on their appearance in the abdomen, adrenal tumours can arise from one adrenal gland (unilateral) or from both adrenal glands (bilateral) and can connect with other organs, including the liver, spleen, pancreas, etc. This connection phenomenon constitutes the most important handicap against adrenal tumour segmentation. Size change, variety of shape, diverse location, and low contrast (similar grey values between the various tissues) are other disadvantages compounding segmentation difficulty. Few studies have considered adrenal tumour segmentation, and no significant improvement has been achieved for unilateral, bilateral, adherent, or noncohesive tumour segmentation. There is also no recognised segmentation pipeline or method for adrenal tumours including different shape, size, or location information. METHODS: This study proposes an adrenal tumour segmentation (ATUS) pipeline designed to eliminate the above disadvantages for adrenal tumour segmentation. ATUS incorporates a number of image methods, including contrast limited adaptive histogram equalization, split and merge based on quadtree decomposition, mean shift segmentation, large grey level eliminator, and region growing. RESULTS: Performance assessment of ATUS was realised on 32 arterial and portal phase computed tomography images using six metrics: dice, jaccard, sensitivity, specificity, accuracy, and structural similarity index. ATUS achieved remarkable segmentation performance, and was not affected by the discussed handicaps, on particularly adherence to other organs, with success rates of 83.06%, 71.44%, 86.44%, 99.66%, 99.43%, and 98.51% for the metrics, respectively, for images including sufficient contrast uptake. CONCLUSIONS: The proposed ATUS system realises detailed adrenal tumour segmentation, and avoids known disadvantages preventing accurate segmentation.

Elimination of white Gaussian noise in arterial phase CT images to bring adrenal tumours into the forefront.

Dynamic Contrast-Enhanced Computed Tomography (DCE-CT) is applied to observe adrenal tumours in detail by utilising from the contrast matter, which generally brings the tumour into the forefront. However, DCE-CT images are generally influenced by noises that occur as the result of the trade-off between radiation doses vs. noise. Herein, this situation constitutes a challenge in the achievement of accurate tumour segmentation. In CT images, most of the noises are similar to Gaussian Noise. In this study, arterial phase CT images containing adrenal tumours are utilised, and elimination of Gaussian Noise is realised by fourteen different techniques reported in literature for the achievement of the best denoising process. In this study, the Block Matching and 3D Filtering (BM3D) algorithm typically achieve reliable Peak Signal-to-Noise Ratios (PSNR) and resolves challenges of similar techniques when addressing different levels of noise. Furthermore, BM3D obtains the best mean PSNR values among the first five techniques. BM3D outperforms to other techniques by obtaining better Total Statistical Success (TSS), CPU time and computation cost. Consequently, it prepares clearer arterial phase CT images for the next step (segmentation of adrenal tumours).

An Efficient Pipeline for Abdomen Segmentation in CT Images.

Computed tomography (CT) scans usually include some disadvantages due to the nature of the imaging procedure, and these handicaps prevent accurate abdomen segmentation. Discontinuous abdomen edges, bed section of CT, patient information, closeness between the edges of the abdomen and CT, poor contrast, and a narrow histogram can be regarded as the most important handicaps that occur in abdominal CT scans. Currently, one or more handicaps can arise and prevent technicians obtaining abdomen images through simple segmentation techniques. In other words, CT scans can include the bed section of CT, a patient's diagnostic information, low-quality abdomen edges, low-level contrast, and narrow histogram, all in one scan. These phenomena constitute a challenge, and an efficient pipeline that is unaffected by handicaps is required. In addition, analysis such as segmentation, feature selection, and classification has meaning for a real-time diagnosis system in cases where the abdomen section is directly used with a specific size. A statistical pipeline is designed in this study that is unaffected by the handicaps mentioned above. Intensity-based approaches, morphological processes, and histogram-based procedures are utilized to design an efficient structure. Performance evaluation is realized in experiments on 58 CT images (16 training, 16 test, and 26 validation) that include the abdomen and one or more disadvantage(s). The first part of the data (16 training images) is used to detect the pipeline's optimum parameters, while the second and third parts are utilized to evaluate and to confirm the segmentation performance. The segmentation results are presented as the means of six performance metrics. Thus, the proposed method achieves remarkable average rates for training/test/validation of 98.95/99.36/99.57% (jaccard), 99.47/99.67/99.79% (dice), 100/99.91/99.91% (sensitivity), 98.47/99.23/99.85% (specificity), 99.38/99.63/99.87% (classification accuracy), and 98.98/99.45/99.66% (precision). In summary, a statistical pipeline performing the task of abdomen segmentation is achieved that is not affected by the disadvantages, and the most detailed abdomen segmentation study is performed for the use before organ and tumor segmentation, feature extraction, and classification.

Application of complex discrete wavelet transform in classification of Doppler signals using complex-valued artificial neural network.

OBJECTIVE: In biomedical signal classification, due to the huge amount of data, to compress the biomedical waveform data is vital. This paper presents two different structures formed using feature extraction algorithms to decrease size of feature set in training and test data. MATERIALS AND METHODS: The proposed structures, named as wavelet transform-complex-valued artificial neural network (WT-CVANN) and complex wavelet transform-complex-valued artificial neural network (CWT-CVANN), use real and complex discrete wavelet transform for feature extraction. The aim of using wavelet transform is to compress data and to reduce training time of network without decreasing accuracy rate. In this study, the presented structures were applied to the problem of classification in carotid arterial Doppler ultrasound signals. Carotid arterial Doppler ultrasound signals were acquired from left carotid arteries of 38 patients and 40 healthy volunteers. The patient group included 22 males and 16 females with an established diagnosis of the early phase of atherosclerosis through coronary or aortofemoropopliteal (lower extremity) angiographies (mean age, 59 years; range, 48-72 years). Healthy volunteers were young non-smokers who seem to not bear any risk of atherosclerosis, including 28 males and 12 females (mean age, 23 years; range, 19-27 years). RESULTS AND CONCLUSION: Sensitivity, specificity and average detection rate were calculated for comparison, after training and test phases of all structures finished. These parameters have demonstrated that training times of CVANN and real-valued artificial neural network (RVANN) were reduced using feature extraction algorithms without decreasing accuracy rate in accordance to our aim.

Complex-valued wavelet artificial neural network for Doppler signals classifying.

OBJECTIVE: In this paper, the new complex-valued wavelet artificial neural network (CVWANN) was proposed for classifying Doppler signals recorded from patients and healthy volunteers. CVWANN was implemented on four different structures (CVWANN-1, -2, -3 and -4). MATERIALS AND METHODS: In this study, carotid arterial Doppler ultrasound signals were acquired from left carotid arteries of 38 patients and 40 healthy volunteers. The patient group had an established diagnosis of the early phase of atherosclerosis through coronary or aortofemoropopliteal angiographies. In implemented structures in this paper, Haar wavelet and Mexican hat wavelet functions were used as real and imaginary parts of activation function on different sequence in hidden layer nodes. CVWANN-1, -2 -3 and -4 were implemented by using Haar-Haar, Mexican hat-Mexican hat, Haar-Mexican hat, Mexican hat-Haar as real-imaginary parts of activation function in hidden layer nodes, respectively. RESULTS AND CONCLUSION: In contrast to CVWANN-2, which reached classification rates of 24.5%, CVWANN-1, -3 and -4 classified 40 healthy and 38 unhealthy subjects for both training and test phases with 100% correct classification rate using leave-one-out cross-validation. These networks have 100% sensitivity, 100% specifity and average detection rate is calculated as 100%. In addition, positive predictive value and negative predictive value were obtained as 100% for these networks. These results shown that CVWANN-1, -3 and -4 succeeded to classify Doppler signals. Moreover, training time and processing complexity were decreased considerable amount by using CVWANN-3. As conclusion, using of Mexican hat wavelet function in real and imaginary parts of hidden layer activation function (CVWANN-2) is not suitable for classifying healthy and unhealthy subjects with high accuracy rate. The cause of unsuitability (obtaining the poor results in CVWANN-2) is lack of harmony between type of activation function in hidden layer and type of input signals in neural network.

Classification of carotid artery Doppler signals in the early phase of atherosclerosis using complex-valued artificial neural network.

In this study, carotid arterial Doppler ultrasound signals were acquired from left carotid arteries of 38 patients and 40 healthy volunteers. The patient group had an established diagnosis of the early phase of atherosclerosis through coronary or aortofemoropopliteal angiographies. Results were classified using complex-valued artificial neural network (CVANN). Principal component analysis (PCA) and fuzzy c-means clustering (FCM) algorithm were used to make a CVANN system more effective. For this aim, before classifying with CVANN, PCA method was used for feature extraction in PCA-CVANN architecture and FCM algorithm was used for data set reduction in FCM-CVANN architecture. Training and test data were selected randomly using 10-fold cross validation. PCA-CVANN and FCM-CVANN architectures classified healthy and unhealthy subjects for training and test data with about 100% correct classification rate. These results shown that PCA-CVANN and FCM-CVANN classified Doppler signals successfully.

A fuzzy clustering neural network architecture for classification of ECG arrhythmias.

Accurate and computationally efficient means of classifying electrocardiography (ECG) arrhythmias has been the subject of considerable research effort in recent years. This study presents a comparative study of the classification accuracy of ECG signals using a well-known neural network architecture named multi-layered perceptron (MLP) with backpropagation training algorithm, and a new fuzzy clustering NN architecture (FCNN) for early diagnosis. The ECG signals are taken from MIT-BIH ECG database, which are used to classify 10 different arrhythmias for training. These are normal sinus rhythm, sinus bradycardia, ventricular tachycardia, sinus arrhythmia, atrial premature contraction, paced beat, right bundle branch block, left bundle branch block, atrial fibrillation and atrial flutter. For testing, the proposed structures were trained by backpropagation algorithm. Both of them tested using experimental ECG records of 92 patients (40 male and 52 female, average age is 39.75 +/- 19.06). The test results suggest that a new proposed FCNN architecture can generalize better than ordinary MLP architecture and also learn better and faster. The advantage of proposed structure is a result of decreasing the number of segments by grouping similar segments in training data with fuzzy c-means clustering.