Artificial Intelligence Algorithms for the Detection of Pathologies Related to Lung Cancer through Image Analysis using Convolutional Neural Networks and Data Augmentation: a systematic mapping of the literature

Lung cancer remains one of the most formidable threats to public health globally, largely because finding it early is often the difference between life and death. The standard method for spotting the disease involves taking detailed three-dimensional pictures of the chest using X-rays, a process known as computed tomography. These images reveal the intricate structures inside the lungs, allowing doctors to look for small, abnormal growths called nodules. However, reading these scans is a demanding task that requires highly trained experts to sift through thousands of images, a bottleneck that can delay diagnosis and limit access to care. In recent years, a new kind of tool has emerged to assist these specialists: artificial intelligence. Specifically, researchers have been developing computer programs inspired by the human brain, known as neural networks, which can learn to recognize patterns in images just as a person does. When these networks are designed specifically for pictures, they are called convolutional neural networks. They work by scanning an image and highlighting the most important features, such as the shape or texture of a nodule, to help determine if it is dangerous.

A recent review of scientific literature, conducted by Pablo Ramirez Amador at the Universidad Abierta Interamericana in Buenos Aires, takes a close look at how these computer systems are currently being used to detect lung cancer. The study examined 96 research articles published between 2015 and the present, searching through major scientific databases to find the most relevant work in the field. The goal was not to test a single new machine, but to map out the entire landscape of existing research, understanding which methods work best and where the biggest hurdles remain. The researchers focused on two specific techniques that have shown promise: transfer learning and data augmentation. Transfer learning allows a computer program that has already learned to recognize general objects, like cats or cars, to be adapted for the specific job of spotting lung nodules, saving time and resources. Data augmentation is a method of creating more training material by taking existing images and slightly altering them—rotating them, changing their size, or adding a bit of visual noise—so the computer sees many more variations of the same thing and learns to be more robust.

The review found that these artificial intelligence tools are indeed capable of offering a highly effective alternative for the early diagnosis of lung cancer, often matching the sensitivity and specificity of human experts. The studies analyzed showed that using pre-trained networks, such as those known as VGG16 or ResNet, combined with data augmentation, leads to high accuracy in identifying non-small cell lung cancer, which is the most common form of the disease. The computer models are able to process the complex data from CT scans and flag potential issues with a level of precision that suggests they could become a vital part of the diagnostic process. However, the paper also makes it clear that this technology is not yet a finished product ready for immediate, widespread use in every hospital. The researchers identified significant challenges that must be solved before these tools can be trusted in real-world clinical settings. One major issue is the lack of standardized data; different hospitals use different machines and protocols, which can confuse the computer models. There are also serious concerns about the privacy of patient information and the need to understand exactly how the computer reaches its conclusions, a quality known as explainability, so that doctors can trust the advice they receive.

Ultimately, the work concludes that while artificial intelligence holds immense potential to improve the care of patients with lung cancer, it is not a magic solution that will replace human doctors. The path forward requires more research to refine these algorithms and new regulations to ensure they are safe, reliable, and ethical. The review suggests that the future of lung cancer detection lies in a partnership where these powerful computer systems act as a second pair of eyes, helping to catch diseases earlier and more accurately, but only if the scientific community addresses the current limitations regarding data quality and model transparency. Until then, the technology remains a powerful promise rather than a fully realized standard of care, waiting for the necessary steps to be taken to bring it safely into the clinic.