Improved detection of focal cortical dysplasia in normal‐appearing FLAIR images using a Bayesian classifier

Abstract

Purpose: Focal cortical dysplasia (FCD) is a malformation of cortical development that often causes pharmacologically intractable epilepsy. However, FCD lesions are frequently characterized by minor structural abnormalities that can easily go unrecognized, making diagnosis difficult. Therefore, many epileptic patients have had pathologically confirmed FCD lesions that appeared normal in pre-surgical fluid-attenuated inversion recovery (FLAIR) magnetic resonance (MR) studies. Such lesions are called "FLAIR-negative." This study aimed to improve the detection of histopathologically verified FCD in a sample of patients without visually appreciable lesions. Methods: The technique first extracts a series of features from a FLAIR image. Then, three naive Bayesian classifiers with probability (NBCP) are trained based on different numbers of feature maps to classify voxels as lesional or healthy voxels and assign the lesions a probability of correct classification. This method classifies the three-dimensional (3D) images of all patients using leave-one-out cross-validation (LOOCV). Finally, the 3D lesion probability map, including epileptogenic lesions, is obtained by removing false-positive voxel outliers using the morphological method. The performance of the NBCP was assessed for quantitative analysis by specificity, accuracy, recall, precision, and Dice coefficient in subject-wise, lesion-wise, and voxel-wise manners. Results: The best detection results were obtained by using four features: cortical thickness, symmetry, K-means, and modified texture energy. There were eight lesions in seven patients. The subjectwise sensitivity of the proposed method was 85.71% (6/7). Seven out of eight lesions were detected, so the lesion-wise sensitivity was 87.50% (7/8). No significant differences in effectiveness were found between automated lesion detection using four features and lesion detection using manual segmentation, as voxels were quantitatively analyzed in terms of specificity (mean AE SD = 99.64 AE 0.13), accuracy (mean AE SD = 99.62 AE 0.14), recall (mean AE SD = 73.27 AE 26.11), precision (mean AE SD = 11.93 AE 8.16), and Dice coefficient (mean AE SD = 22.82 AE 15.57). Conclusion: We developed a novel automatic voxel-based method to improve the detection of FCD FLAIR-negative lesions. To the best of our knowledge, this study is the first to detect FCD lesions that appear normal in pre-surgical 3D high-resolution FLAIR images alone with a limited number of radiomics features. We optimized the algorithm and selected the best prior probability to improve the detection. For non-temporal lobe epilepsy (non-TLE) patients, lesions could be accurately located, although there were still false-positive areas.

A precise bare simulation approach to the minimization of some distances. Foundations

“…There is a vast literature on very recent applications of the Bhattacharyya coefficient, for instance it appears exemplarily in Peng & Li [289] for object tracking from successive video frames, Ayed et al [26] for efficient graph cut algorithms, Patra et al [287] for collaborative filtering in sparse data, El Merabet et al [119] for region classification in intelligent transport systems in order to compensate the lack of performance of Global Navigation Satellites Systems, Chiu et al [86] for the design of interactive mobile augmented reality systems, Noh et al [274] for dimension reduction in interacting fluid flow models, Bai et al [29] for material defect detection through ultrasonic array imaging, Dixit & Jain [115] for the design of recommender systems on highly sparse context aware datasets, Guan et al [143] for visible light positioning methods based on image sensors, Lin et al [220] for probabilistic representation of color image pixels, Chen et al [80] for distributed compressive video sensing, Jain et al [162] for the enhancement of multistage user-based collaborative filtering in recommendation systems, Pascuzzo et al [285] for brain-diffusion-MRI based early diagnosis of the sporadic Creutzfeldt-Jakob disease, Sun et al [351] for the design of automatic detection methods multitemporal (e.g. landslide) point clouds, Valpione et al [377] for the investigation of T cell dynamics in immunotherapy, Wang et al [387] for the tracking and prediction of downbursts from meteorological data, Xu et al [403] for adaptive distributed compressed video sensing for coal mine monitoring, Zhao et al [424] for the shared sparse machine learning of the affective content of images, Chen et al [82] for image segmentation and domain partitioning, De Oliveira et al [105] for the prediction of cell-penetrating peptides, Eshaghi et al [122] for the identification of multiple sclerosis subtypes through machine learning of brain MRI scans, Feng et al [125] for improvements of MRI-based detection of epilepsy-causing cortical malformations, Hanli et al [153] for designing pilot protection schemes for transmission lines, Jiang et al [170] for flow-assisted visual tracking through event cameras, Lysiak & Szmajda …”

“…Let us finally mention that an important special case of a minimization problem (125) to ( 129) is -the integer programming formulation of -the omnipresent (asymmetric) traveling salesman problem (TSP) with possible side constraints 27 . There, one has K cities and the cost of traveling from city i to city j = i is given by c ij > 0.…”

A Multi-Task Convolutional Neural Network for Lesion Region Segmentation and Classification of Non-Small Cell Lung Carcinoma

“…The lesion segmentation task aims to detect tumor location and boundaries, and the tumor classification task aims to identify tumor histological subtypes. In previous studies, many traditional machine learning methods were presented, such as the combination of a probabilistic neural network and support vector machines (PNN-SVM) [ 3 , 4 ], the Bayesian classifier [ 5 , 6 ] and the neural-like structure of successive geometric transformations model (SGTM) [ 7 , 8 , 9 , 10 ] for tumor classification and Gibbs random field [ 11 ], fuzzy C-means [ 12 ] and Wavelet Analysis [ 13 ] for segmentation. These methods highly rely on hand-crafted feature engineering and are unable to learn deep representations from visual levels.…”

Vocal cord lesions classification based on deep convolutional neural network and transfer learning

“…Fortunately, computer-aided diagnosis (CAD) has the potential to boost accuracy and efficiency, providing the human expert with an objective complementary measure. [6][7][8][9] Deep learning, particularly deep convolutional neural network (DCNN), has achieved promising results in many fields. It has obtained many encouraging progresses in auxiliary diagnoses, such as skin cancers detection based on dermoscopy images, 10 lung diseases diagnoses based on histopathology images, 11 diabetic retinopathy diagnosis based on retinal fundus images, 12 and bacteria identification based on fluorescence microscopy images.…”

“…Stemming from this subjective way, examination results vary from doctors with uneven experience, and cases may be misjudged. Fortunately, computer‐aided diagnosis (CAD) has the potential to boost accuracy and efficiency, providing the human expert with an objective complementary measure 6–9 …”