Dr. Mian Muhammad Sadiq Fareed, a Portugal-based Pakistani researcher specializing in artificial intelligence and medical imaging, has introduced a groundbreaking deep learning model that may significantly improve the early detection of Alzheimer’s disease. His newly developed system, known as ADD-Net, uses MRI scans to identify subtle brain changes associated with different stages of dementia, offering a powerful new approach to diagnosing the condition long before severe symptoms appear. Alzheimer’s disease remains one of the world’s most challenging neurological disorders, with early diagnosis often missed due to the limitations of manual MRI interpretation. Dr. Fareed explains that deep learning can analyze brain images at a level of precision that surpasses what the human eye can detect, making it possible to catch structural abnormalities at the earliest possible stage.
In his research, ADD-Net was trained to detect four categories of cognitive health: normal aging, very mild dementia, mild dementia, and moderate dementia. By analyzing thousands of MRI images, the model learns patterns of brain atrophy and other deviations that correspond to these stages. He emphasizes that artificial intelligence allows researchers to uncover deeper structural signals that remain hidden in traditional methods, offering a reliable, faster, and more objective diagnostic tool.
One of the biggest challenges in medical artificial intelligence is the imbalance in datasets, where certain disease categories contain only a few samples. This often leads to inaccurate models. To address this issue, Dr. Fareed used a specialized technique called SMOTETomek, which generates high-quality synthetic MRI images to balance under-represented classes. He notes that without solving the imbalance problem, no model can be expected to perform reliably in clinical environments.
ADD-Net was tested against well-known deep learning architectures such as DenseNet169, VGG19, and InceptionResNet-V2. The new model outperformed them all, achieving 98.63% accuracy, 99.76% AUC, and exceptionally strong precision and recall scores. These results place ADD-Net among the most accurate Alzheimer’s detection models reported so far and demonstrate its potential for real-world medical deployment.
Another important aspect of Dr. Fareed’s work is the use of Grad-CAM heatmaps, a technique that visually highlights the specific regions of the brain the model focuses on when making predictions. This transparency is crucial for clinical acceptance, allowing doctors to understand why the system identifies a patient as having mild or moderate dementia and helping them trust AI-assisted diagnosis. These visual insights also deepen scientific understanding of how structural damage progresses in Alzheimer’s disease.
Dr. Fareed believes the model could support neurologists by reducing diagnostic delays and enabling earlier intervention. He aims to continue refining the model and expanding it with larger and more diverse datasets, hoping to integrate transfer learning and more advanced neuro-imaging techniques. Committed to advancing medical AI, he views ADD-Net as an important step toward more precise, accessible, and effective diagnostic tools for patients and families affected by Alzheimer’s disease.