Othmane Daanouni

Accurate classification of brain tumors (BT) from magnetic resonance imaging MRI remains challenging due to noise, variability in tumor morphology, and inherent uncertainty in medical images, which often degrade the performance of deep learning models under real world conditions. To address this limitation, we propose a novel hybrid framework, NS Xception, that integrates neutrosophic based image enhancement with a deep convolutional neural network for noise resilient brain tumor classification. In this approach, MRI images are first transformed into the neutrosophic domain, where each pixel is represented by truth, indeterminacy, and falsity components, enabling explicit modeling of uncertainty. Enhancement operations are then applied to suppress noise and highlight diagnostically relevant structures, and the refined truth component is used as input to a customized Xception model for multiclass classification. The proposed method is evaluated on a multi source MRI dataset comprising four classes, glioma, meningioma, pituitary tumor, and no tumor, and achieves a classification accuracy of 97.94 % under clean conditions, outperforming several state of the art approaches. Importantly, under noisy conditions, it maintains significantly higher performance compared to conventional denoising techniques such as Gaussian and bilateral filtering, demonstrating strong robustness to image degradation. These results indicate that incorporating uncertainty aware representations through neutrosophic theory can substantially enhance the stability and reliability of deep learning models in medical image analysis, making the proposed framework a promising solution for robust computer aided diagnosis in practical clinical environments.

Convolution Neural Network (CNN) models have gained ground in research activities particularly in medical images used for Diabetes Retinopathy (DR) detection. X-ray, MRI, and CT scans have all been used to validate CNN models, with classification accuracy generally reaching that of trained doctors. It is mandatory to evaluate the strength of CNN models used in medical tasks against adversarial attacks especially in healthcare; that is to say, the security of such models is becoming extremely relevant to the diagnosis as this latter will guide high-stakes decision-making. However, little study has been conducted to better comprehend this issue. This paper focuses on MobileNet CNN architecture in order to investigate its vulnerability against fast gradient sign methods (FGSM) adversarial attacks. For this end, a Neural Structure Learning (NSL) and a Multi-Head Attention (MHA) have been used to effectively reduce the vulnerability against attack by end-to-end CNN training with adversarial neighbors that produce adversarial perturbations on optical coherence tomography (OCT) images. With suggested model NSL-MHA-CNN, there has been an ability to maintain model performance on adversarial attack without increasing cost of training. Through theoretical assistance and empirical validation, it was possible to examine the stability of MobileNet architecture and demonstrate its susceptibility, particularly to adversarial attack. The experiments in this paper show that indiscernible degrees of perturbation ε < 0.01 were sufficient to cause a task failure resulting to misclassification in majority of the time. Moreover, empirical simulation shows that the proposed approach advanced in this paper can be an effective method to defense against adversarial attack at level of CNN model testing.