Deep learning requires a large amount of data to perform well. However, the field of medical image analysis suffers from a lack of sufficient data for training deep learning models. Moreover, medical images require manual labeling, usually provided by human annotators coming from various backgrounds. More importantly, the annotation process is time-consuming, expensive, and prone to errors. Transfer learning was introduced to reduce the need for the annotation process by transferring the deep learning models with knowledge from a previous task and then by fine-tuning them on a relatively small dataset of the current task. Most of the methods of medical image classification employ transfer learning from pretrained models, e.g., ImageNet, which has been proven to be ineffective. This is due to the mismatch in learned features between the natural image, e.g., ImageNet, and medical images. Additionally, it results in the utilization of deeply elaborated models. In this paper, we propose a novel transfer learning approach to overcome the previous drawbacks by means of training the deep learning model on large unlabeled medical image datasets and by next transferring the knowledge to train the deep learning model on the small amount of labeled medical images. Additionally, we propose a new deep convolutional neural network (DCNN) model that combines recent advancements in the field. We conducted several experiments on two challenging medical imaging scenarios dealing with skin and breast cancer classification tasks. According to the reported results, it has been empirically proven that the proposed approach can significantly improve the performance of both classification scenarios. In terms of skin cancer, the proposed model achieved an F1-score value of 89.09% when trained from scratch and 98.53% with the proposed approach. Secondly, it achieved an accuracy value of 85.29% and 97.51%, respectively, when trained from scratch and using the proposed approach in the case of the breast cancer scenario. Finally, we concluded that our method can possibly be applied to many medical imaging problems in which a substantial amount of unlabeled image data is available and the labeled image data is limited. Moreover, it can be utilized to improve the performance of medical imaging tasks in the same domain. To do so, we used the pretrained skin cancer model to train on feet skin to classify them into two classes—either normal or abnormal (diabetic foot ulcer (DFU)). It achieved an F1-score value of 86.0% when trained from scratch, 96.25% using transfer learning, and 99.25% using double-transfer learning.

The spectrum has increasingly become occupied by various wireless technologies. For this reason, the spectrum has become a scarce resource. In prior work, the authors have addressed the spectrum sensing problem by using multi-input and multi-output (MIMO) in cognitive radio systems. We considered the detection and estimation framework for MIMO cognitive network where the noise covariance matrix is unknown with perfect channel state information. In this study, we propose a generalized likelihood ratio test (GLRT) for the spectrum sensing problem in cognitive radio where the noise covariance matrix is unknown with non-perfect channel state information. Two scenarios are examined in this study: (i) in the first scenario, the sub-optimal solution of the worst case of the system’s performance is considered; (ii) in the second scenario, we present a robust detector for the MIMO spectrum sensing problem. For both scenarios, the Bayesian approach with a generalized likelihood ratio test based on the binary hypothesis problem is used. From the results, it can be seen that our approach provides the best performance in the spectrum sensing problem under specified assumptions. The simulation results also demonstrate that our approach significantly outperforms other state-of-the-art spectrum sensing detectors when the channel uncertainty is addressed.

Achieving high quality of service (QOS) at the end‐users while maintaining the low interference power at the primary users is the main goal in underlay cognitive radio networks. This goal becomes a more difficult task in the designing of beamforming vectors where all channels state information (CSIs) in the secondary network, as well as the interference CSIs, are uncertain. This task is addressed in this study using an iterative optimization technique. In this technique, the original CSI problem, which is difficult to solve as a single optimization problem, is instead separated into two sub‐problems. The first subproblem is the interference power problem, which can be solved either sub‐optimally or optimally using Lagrange duality. The second sub‐problem is the QOS problem, which can be solved either sub‐optimally or robustly using non‐monotone spectral projected gradient method. The two sub‐problem solutions are then recombined into a single problem to extract beamforming vectors. Two methods are invoked to extract the beamforming vectors: either the successive convex approximation (SCA) method or the bisection search method. Theoretical analysis and simulation results indicate that the two methods can offer a tradeoff between better QOS (using bisection search method) or less computational complexity (using SCA method)

The detection of an active primary user (PU) by a secondary base station (SBS) is essential for an interweave cognitive radio (CR). This work proposes a Bayesian method for model order selection to the detection problem in a multi-PU single-SBS CR network, where the PUs and the SBS are equipped with multiple antennas. Each PU transmits separate known beacon sequences when they are active but the multiple-input multiple-output (MIMO) channels from the PUs to SBS are not known. We select a Bayesian model among potential models of the received beacon sequences to decide the number of active PUs. It is accomplished by imposing Zellner’s g-prior on the unknown MIMO channels, obtaining the Bayes factor (BF) between an alternative hypothesis and the null hypothesis, and evaluating the posterior probability from the BFs as the test statistic. The resulting BF has an analytical form in terms of the hyperparameter g from the prior. We develop three methods to address the unknown g. The first method assigns the inverse-gamma hyperprior to g and uses the expectation of BF over g instead. The second estimates g locally which is different for each data model and the third obtains the g estimate globally which is common to all candidate models. We conduct rigorous analysis for the proposed Bayesian selectors and determine the suitable ranges for the shape and scale parameters of the g-prior. In addition, analysis shows that they are consistent in fulfilling the information and model selection criteria. Numerical results reveal that the three proposed selectors have comparable performance and outperform the baselines Akaike information criterion (AIC), Bayesian information criterion (BIC), extended BIC (EBIC), and false discovery rate (FDR) at various settings of the signal-to-noise ratio (SNR), the length of beacon sequences, and the number of SBS antennas.