Retinal diseases are among the leading causes of vision impairment worldwide, and early detection is essential for enabling personalized treatments and preventing irreversible vision loss. In this paper, we propose a method aimed to identify and localize retinal conditions, i.e., Age-Related Macular Degeneration, Diabetic Retinopathy, and Choroidal Neovascularization, using explainable deep learning. For this purpose, we consider seven fine-tuned convolutional neural networks: MobileNet, LeNet, StandardCNN, CustomCNN, DenseNet, Inception, and EfficientNet. Moreover, we develop a novel architecture i.e., NeoNet, specifically designed for the detection of retinal diseases, achieving an accuracy of 99.5%. Furthermore, with the aim to provide explaianability behind the model decision, we highlight the most critical regions within retinal images influencing the predictions of the model. The obtained results show the ability of the model to detect pathological features, thereby supporting earlier and more accurate diagnosis of retinal diseases.

Background:Generative Adversarial Networks (GANs), thanks to their great versatility, have a plethora of applications in biomedical imaging with the goal of simulating complex pathological conditions and creating clinical data used for training advanced machine learning models. The ability to generate high-quality synthetic clinical data not only addresses issues related to the scarcity of annotated bioimages but also supports the continuous improvement of diagnostic tools. Method: We propose a two-step method aimed to detect whether a bioimage can be considered fake or real. The first step is related to bioimage generation using a Deep Convolutional GAN, while the second step involves the training and testing of a set of machine learning models aimed to distinguish between real and generated bioimages. Results: We evaluate our approach by exploiting six different datasets. We observe notable results, demonstrating the ability of Deep Convolutional GAN to generate realistic synthetic images for some specific bioimages. However, for other bioimages, the accuracy does not align with the expected trend, indicating challenges in generating images that closely resemble real ones. Conclusions: This study highlights both the potential and limitations of GAN in generating realistic bioimages. Future work will focus on improving generation quality and detection accuracy across different datasets.

Background: Oral cavity tumors are often diagnosed at advanced stages due to non-specific early symptoms and limitations in imaging sensitivity. This study proposes a novel methodology combining Radiomics and Model Checking to improve diagnosis and prognosis. Methods: A retrospective dataset of 18 patients (12 with oral squamous cell carcinoma and 6 healthy controls) who underwent contrast-enhanced MRI was analyzed. Radiomic features were extracted and selected, then encoded into formal models. A two-tier Model Checking approach was applied: (i) classification of healthy vs. pathological patients and (ii) prediction of treatment response. Results: The proposed method achieved a diagnostic accuracy of 93% and a prognostic accuracy of 75%. The approach demonstrated robustness even with a limited dataset, outperforming traditional data-driven methods in small-sample settings. Conclusions: The integration of Radiomics with Model Checking provides an explainable and effective tool for early detection and prognosis of oral cavity tumors. This approach may support clinicians as a decision-making aid, particularly in data-scarce scenarios.