A machine learning approach for prediction of auditory brain stem response in patients after head-and-neck radiation therapy.

Objective: The present study aimed to assess machine learning (ML) models according to radiomic features to predict ototoxicity using auditory brain stem responses (ABRs) in patients with radiation therapy (RT) for head-and-neck cancers. Materials and Methods: The ABR test was performed on 50 patients having head-and-neck RT. Radiomic features were extracted from the brain stem in computed tomography images to generate a radiomic signature. Moreover, accuracy, sensitivity, specificity, the area under the curve, and mean cross-validation were used to evaluate six different ML models. Results: Out of 50 patients, 21 participants experienced ototoxicity. Furthermore, 140 radiomic features were extracted from the segmented area. Among the six ML models, the Random Forest method with 77% accuracy provided the best result. Conclusion: According to the ML approach, we showed the relatively high prediction power of the radiomic features in radiation-induced ototoxicity. To better predict the outcomes, future studies on a larger number of participants are recommended.

Radiomics analysis on CT images for prediction of radiation-induced kidney damage by machine learning models.

INTRODUCTION: We aimed to assess the power of radiomic features based on computed tomography to predict risk of chronic kidney disease in patients undergoing radiation therapy of abdominal cancers. METHODS: 50 patients were evaluated for chronic kidney disease 12 months after completion of abdominal radiation therapy. At the first step, the region of interest was automatically extracted using deep learning models in computed tomography images. Afterward, a combination of radiomic and clinical features was extracted from the region of interest to build a radiomic signature. Finally, six popular classifiers, including Bernoulli Naive Bayes, Decision Tree, Gradient Boosting Decision Trees, K-Nearest Neighbor, Random Forest, and Support Vector Machine, were used to predict chronic kidney disease. Evaluation criteria were as follows: accuracy, sensitivity, specificity, and area under the ROC curve. RESULTS: Most of the patients (58%) experienced chronic kidney disease. A total of 140 radiomic features were extracted from the segmented area. Among the six classifiers, Random Forest performed best with the accuracy and AUC of 94% and 0.99, respectively. CONCLUSION: Based on the quantitative results, we showed that a combination of radiomic and clinical features could predict chronic kidney radiation toxicities. The effect of factors such as renal radiation dose, irradiated renal volume, and urine volume 24-h on CKD was proved in this study.

Automated thyroid nodule detection from ultrasound imaging using deep convolutional neural networks.

Thyroid cancer is the most common endocrine cancer and its incidence has continuously increased worldwide. In this paper, we focus on the challenging problem of nodule detection from ultrasound scans. In current clinical practice, this task is performed manually, which is tedious, subjective and highly depends on the clinical experience of radiologists. We propose a novel deep neural network architecture with carefully designed loss function regularization, and network hyperparameters to perform nodule detection without complex post-processing refinement steps. The local training and validation datasets consist of 2461 and 820 ultrasound frames acquired from 60 and 20 patients with a high degree of variability, respectively. The core of the proposed method is a deep learning framework based on multi-task model Mask R-CNN. We have developed a loss function with regularization that prioritizes detection over segmentation. Validation was conducted for 821 ultrasound frames from 20 patients. The proposed model can detect various types of thyroid nodules. The experimental results indicate that our proposed method is effective in thyroid nodule detection. Comparisons with the results by Faster R-CNN and conventional Mask R-CNN demonstrate that the proposed model outperforms the prior state-of-the-art detection methods.

A novel image-based retrieval system for characterization of maxillofacial lesions in cone beam CT images.

PURPOSE: The objective of medical content-based image retrieval (CBIR) is to assist clinicians in decision making by retrieving the most similar cases to a given query image from a large database. Herein, a new method for content-based image retrieval of cone beam CT (CBCT) scans is presented. METHODS: The introduced framework consists of two main phases: training database construction and querying. The goal of the training phase is database construction, which consists of three main steps. First, automatic segmentation of lesions using 3D symmetry analysis is performed. Embedding the prior shape knowledge of the 3D symmetry characteristics of the healthy human head structure increases the accuracy of automatic segmentation. Then, spatial pyramid matching is used for feature extraction, and the relative importance of each feature is learned using classifiers. RESULTS: The method was applied to a dataset of 1145 volumetric CBCT images with four classes of maxillofacial lesions. A symmetry-based analysis model for automatic lesion segmentation was evaluated using similarity measures. Mean Dice coefficients of 0.89, 0.85, 0.92, and 0.87 were achieved for maxillary sinus perforation, radiolucent lesion, unerupted tooth, and root fracture classes, respectively. Moreover, the execution time of automatic segmentation was reduced to 3 min per case. The performance of the proposed search engine was evaluated using mean average precision and normalized discounted cumulative gain. A mean average retrieval accuracy and normalized discounted cumulative gain of 0.90 and 0.92, respectively, were achieved. CONCLUSION: Quantitative results show that the proposed approach is more effective than previous methods in the literature, and it can facilitate the introduction of CBIR in clinical CBCT applications.

Automated classification of maxillofacial cysts in cone beam CT images using contourlet transformation and Spherical Harmonics.

BACKGROUND AND OBJECTIVE: Accurate detection of maxillofacial cysts is an essential step for diagnosis, monitoring and planning therapeutic intervention. Cysts can be of various sizes and shapes and existing detection methods lead to poor results. Customizing automatic detection systems to gain sufficient accuracy in clinical practice is highly challenging. For this purpose, integrating the engineering knowledge in efficient feature extraction is essential. METHODS: This paper presents a novel framework for maxillofacial cysts detection. A hybrid methodology based on surface and texture information is introduced. The proposed approach consists of three main steps as follows: At first, each cystic lesion is segmented with high accuracy. Then, in the second and third steps, feature extraction and classification are performed. Contourlet and SPHARM coefficients are utilized as texture and shape features which are fed into the classifier. Two different classifiers are used in this study, i.e. support vector machine and sparse discriminant analysis. Generally SPHARM coefficients are estimated by the iterative residual fitting (IRF) algorithm which is based on stepwise regression method. In order to improve the accuracy of IRF estimation, a method based on extra orthogonalization is employed to reduce linear dependency. We have utilized a ground-truth dataset consisting of cone beam CT images of 96 patients, belonging to three maxillofacial cyst categories: radicular cyst, dentigerous cyst and keratocystic odontogenic tumor. RESULTS: Using orthogonalized SPHARM, residual sum of squares is decreased which leads to a more accurate estimation. Analysis of the results based on statistical measures such as specificity, sensitivity, positive predictive value and negative predictive value is reported. The classification rate of 96.48% is achieved using sparse discriminant analysis and orthogonalized SPHARM features. Classification accuracy at least improved by 8.94% with respect to conventional features. CONCLUSIONS: This study demonstrated that our proposed methodology can improve the computer assisted diagnosis (CAD) performance by incorporating more discriminative features. Using orthogonalized SPHARM is promising in computerized cyst detection and may have a significant impact in future CAD systems.

Automatic segmentation of mandibular canal in cone beam CT images using conditional statistical shape model and fast marching.

PURPOSE: Accurate segmentation of the mandibular canal in cone beam CT data is a prerequisite for implant surgical planning. In this article, a new segmentation method based on the combination of anatomical and statistical information is presented to segment mandibular canal in CBCT scans. METHODS: Generally, embedding shape information in segmentation models is challenging. The proposed approach consists of three main steps as follows: At first, a method based on low-rank decomposition is proposed for preprocessing. Then, a conditional statistical shape model is trained, and mandibular bone is segmented with high accuracy. In the final stage, fast marching with a new speed function is utilized to find the optimal path between mandibular and mental foramen. Fast marching tries to find the darkest tunnel close to the initial segmentation of the canal, which was obtained with conditional SSM model. In this regard, localization of mandibular canal is performed more accurately. RESULTS: The method is applied to the identification of mandibular canal in 120 sets of CBCT images. Conditional statistical model is evaluated by calculating the compactness capacity, specificity and generalization ability measures. The capability of the proposed model is evaluated in the segmentation of mandibular bone and canal. The framework is effective in noisy scans and is able to detect canal in cases with mild bone resorption. CONCLUSION: Quantitative analysis of the results shows that the method performed better than two other recent methods in the literature. Experimental results demonstrate that the proposed framework is effective and can be used in computer-guided dental implant surgery.

Automatic segmentation of maxillofacial cysts in cone beam CT images.

Accurate segmentation of cysts and tumors is an essential step for diagnosis, monitoring and planning therapeutic intervention. This task is usually done manually, however manual identification and segmentation is tedious. In this paper, an automatic method based on asymmetry analysis is proposed which is general enough to segment various types of jaw cysts. The key observation underlying this approach is that normal head and face structure is roughly symmetric with respect to midsagittal plane: the left part and the right part can be divided equally by an axis of symmetry. Cysts and tumors typically disturb this symmetry. The proposed approach consists of three main steps as follows: At first, diffusion filtering is used for preprocessing and symmetric axis is detected. Then, each image is divided into two parts. In the second stage, free form deformation (FFD) is used to correct slight displacement of corresponding pixels of the left part and a reflected copy of the right part. In the final stage, intensity differences are analyzed and a number of constraints are enforced to remove false positive regions. The proposed method has been validated on 97 Cone Beam Computed Tomography (CBCT) sets containing various jaw cysts which were collected from various image acquisition centers. Validation is performed using three similarity indicators (Jaccard index, Dice's coefficient and Hausdorff distance). The mean Dice's coefficient of 0.83, 0.87 and 0.80 is achieved for Radicular, Dentigerous and KCOT classes, respectively. For most of the experiments done, we achieved high true positive (TP). This means that a large number of cyst pixels are correctly classified. Quantitative results of automatic segmentation show that the proposed method is more effective than one of the recent methods in the literature.

Using phase space reconstruction for patient independent heartbeat classification in comparison with some benchmark methods.

Many methods for automatic heartbeat classification have been applied and reported in literature, but relatively few of them concerned with patient independent classification because of the less significant results compared to patient dependent ones. In this work, using phase space reconstruction in order to classify five heartbeat types can fill this gap to some extent. In the first and second method, Reconstructed phase space (RPS) is modeled by the Gaussian mixture model (GMM) and bins, respectively, and then classified by classic Bayesian classifier. In the third method, RPS is directly used to train predictor time-delayed neural networks (TDNN) and classified based on minimum prediction error. All three methods highly outperform the results reported before, for patient independent heartbeat classification. The best result is achieved using GMM-Bayes method with 92.5% classification accuracy.