AI in Disease Detection

                    AI is Revolutionizing Disease Detection

Within the last couple of years, “Artificial Intelligence” has become a household name. It has evolved from a taboo concept to a central entity in modern medicine. The best kind of medicine is disease prevention, but humans are not equipped to predict the unreliability of the human anatomy and the disease/immune system. Therefore, lately, various intelligence models have been programmed and trained to identify complex medical patterns in imaging and test results that even the best physicians may miss. As demand for healthcare is through the roof and physicians are spread too thin, AI is becoming a critical partner in clinical decision-making. 

As a result, an auspicious development in the AI realm is the use of artificial neural networks (ANNs). ANNs are an adaptive type of intelligence modeled after the human brain. They are fascinating because they can process vast volumes of imaging data in minutes. Not only are they incredibly accurate in identifying early signs of cancer or respiratory abnormalities, but ANNs also offer speed and scalability.  However, when a new system gains popularity, physicians and patients become skeptical about how it can achieve its claimed accuracy and whether that accuracy is absolute. Is it true that the ANNs are “seeing” the imaging in the same way a radiologist would? 

A 2023 study by Alsharif et al., published in Bioengineering, addresses this question. The author and his partners review how ANNs played a role in diagnosing COVID-19 while noting an advancement called ConXNet. This particular program was designed to detect COVID-19 by simply processing chest X-rays. When tested, it was found that ConXNEt achieved 97.8% accuracy in its diagnosis. The author continued to explain what makes ConXNet stand out from its competitors. It was determined that its advantages lie in its thoughtful training process. The ConXNet program utilizes dropout layers, batch normalization, and convolutional blocks, which help reduce overfitting and improve generalization. Furthermore, another essential ConXNEt trait is its high success rate when tested with extensive image pools and more diverse scans. This means that ConXNEt applies to real-world situations.

Additionally, the pinnacle of this program's innovative genius lies in its ability to accurately interpret the visuals given to it. When a physician deciphers an image, they have to be able to locate even the most minor differences in texture, shape, or color- because this can mean the difference between health and life-threatening disease. However, how does this neural network see these slight differences and distinguish between healthy and diseased? This is where research in The Journal of Vision offers a valuable perspective. Researchers from Loyola University found that ANNs have difficulty visualizing images from unusual angles, relying instead on relations and 3D visual cues. This means that while ConXNet is incredibly impressive, it still can improve how it interprets depth, spatial orientation, and shape changes. 

As ConXNEt continues to break barriers and set the foundation for ANNs in medical imaging, it can still be improved. Recognizing diseases isn't just about detecting differences in color and texture but understanding what differences in shape could mean. AI needs to be able to see images the same way a physician would; one day, it may even surpass them in every way. 

References 

Azeem, M.; Javaid, S.; Khalil, R.A.; Fahim, H.; Althobaiti, T.; Alsharif, N.; Saeed, N. Neural Networks for the Detection of COVID-19 and Other Diseases: Prospects and Challenges. Bioengineering 2023, 10, 850. https://doi.org/10.3390/ bioengineering10070850

Cutler, M; Baumel, L; Tocco, J; Friebel, W; Thiruvathukal, G; Baker, N. Beyond the Contour: How 3D Cues Enhance Object Recogntion in Humans and Neural Networks: Journal of Vision (20??),?,1-?.file:///Users/yoliestarck/Downloads/Cutler%20&%20Baumel%202025%20(Pre-Publication)%20(3).pdf (not yet published).