PulmoDiagAI: An Intelligent Medical Diagnosis System for Automated Pulmonary Pathology Classification using Deep Learning

The human lung is a delicate, branching network of airways and sacs, designed to exchange oxygen for carbon dioxide with every breath. When this system falters, the consequences can be swift and severe. Two of the most pressing threats to respiratory health today are lung cancer and the viral infection known as COVID-19. Both conditions often hide in plain sight within the body, revealing themselves only through the subtle shadows and textures visible on medical images like X-rays and CT scans. For decades, the task of interpreting these images has fallen to radiologists, highly trained specialists who must scan thousands of pictures, looking for the faint signs of disease. This process is vital, but it is also exhausting and prone to human error, especially when doctors are overwhelmed by a high volume of patients or when the signs of illness are ambiguous.

In recent years, a new tool has emerged to assist these medical experts: artificial intelligence. Specifically, a branch of AI called deep learning allows computers to learn how to recognize patterns in images, much like a human brain does, but by processing millions of examples at once. Researchers have been training these digital systems to spot the tell-tale marks of pneumonia, tumors, and viral infections. The goal is not to replace the doctor, but to provide a second pair of eyes that never gets tired, never misses a detail due to fatigue, and can offer a rapid, consistent opinion on what a patient is facing. The question remains, however, which specific type of digital brain works best for which specific disease, and can these complex systems be made simple enough for a hospital to use every day?

A team of researchers from Mouloud Mammeri University in Algeria has tackled these questions with a new system they call PulmoDiagAI. Their work focuses on creating an intelligent tool that can automatically sort through chest images to diagnose two distinct threats: lung cancer and COVID-19. To build this system, they did not invent a new type of computer brain from scratch. Instead, they tested five different existing architectures—pre-built deep learning models that have already been trained on vast libraries of general images. These models, known by names like VGG19, ResNet50, and EfficientNet-B3, were adapted to look specifically at the human chest. The researchers fed these models thousands of medical images, teaching them to distinguish between healthy lungs, benign (non-cancerous) growths, malignant (cancerous) tumors, and viral infections.

The process began with a careful preparation of the images themselves. Medical scans often come with variations in brightness, noise, or contrast that can confuse a computer. The team applied a series of digital enhancements to every picture, converting them to grayscale, removing static-like noise, and sharpening the contrast to make the edges of the lungs and any potential abnormalities stand out clearly. This step was crucial, as it ensured that the artificial intelligence was looking at the clearest possible version of the patient's anatomy before it began its diagnosis. Once the images were ready, the researchers let the five different models compete against each other, testing their ability to correctly identify the condition in the picture.

The results revealed that no single model was perfect for every job; the best choice depended entirely on the disease being investigated. When it came to identifying COVID-19 from chest X-rays, one model, EfficientNet-B3, rose above the rest. It correctly identified the virus in 97% of cases and demonstrated an exceptional ability to distinguish between infected patients, those with other types of pneumonia, and those with healthy lungs. This model proved so reliable that it could almost perfectly separate the three groups, making it the ideal candidate for detecting viral outbreaks.

However, the story changed when the team turned their attention to lung cancer. For this more complex task, which involved analyzing CT scans to find tumors, a different model took the lead. MobileNetV3, a design originally created to be lightweight and fast for mobile devices, achieved a stunning 99% accuracy. Even more impressively, both MobileNetV3 and another model called ResNet50 managed to find every single case of malignant cancer in their test sets. They did not miss a single dangerous tumor. This is a critical finding, as in medical screening, missing a cancer diagnosis is far more dangerous than raising a false alarm. While other models sometimes struggled to tell the difference between a harmless growth and a healthy lung, these top performers maintained a sharp distinction, ensuring that no life-threatening condition slipped through the cracks.

The researchers did not stop at simply finding the best numbers. They understood that a powerful algorithm is useless if a doctor cannot easily use it. To bridge this gap, they built a simple, interactive computer program that anyone could operate. This interface allows a clinician to upload a chest image, select whether they are looking for cancer or COVID-19, and receive an instant answer. The system does not just say "cancer" or "no cancer"; it provides a confidence score, a percentage that tells the doctor how sure the computer is about its finding. This transparency helps the medical professional decide how much weight to give the computer's opinion, turning a complex mathematical calculation into a practical, real-time assistant.

The study confirms that deep learning can be a powerful ally in the fight against respiratory diseases, but it also highlights that there is no one-size-fits-all solution. The best tool for spotting a virus is not necessarily the best for finding a tumor. By carefully matching the right digital brain to the right medical task, and by wrapping that technology in a user-friendly interface, the researchers have created a system that is ready for the real world. PulmoDiagAI stands as a demonstration that when advanced artificial intelligence is paired with careful medical preparation, it can offer a reliable, rapid, and accessible way to support doctors in saving lives.