Foreword
This post is the first in a series on the use of artificial intelligence and machine learning in contemporary medicine.
The Advent of Machine Learning in Radiology
In today’s day and age, machine learning is a ubiquitous technology, seemingly impacting every aspect of our lives and each sector of society. The field of medicine, and radiology in particular, is no exception to this notion. The fusion of machine learning and radiology promises enhanced diagnostic accuracy, improved patient outcomes, and the optimized delivery of healthcare. Recent advancements in machine learning algorithms and data availability have ushered in a new era of precision, accuracy, and efficiency in radiology, revolutionizing the field.
From the dawn of radiology until now, the interpretation of medical images has relied on the expertise of radiologists (i.e., human beings) to provide diagnoses. However, such manual analyses can be made challenging by the great volume and complexity of data that is being continually generated. This is where machine learning can really shine, empowering algorithms (i.e., computers) to learn from data without being explicitly programmed.
The development and recent advancements of deep learning algorithms (especially convolutional neural networks, which are great for image recognition) have been a huge breakthrough in radiology. These algorithms are capable of poring over huge repositories of medical images with speeds and accuracies well beyond those of humans, which can be of great help to radiologists in detecting subtle anomalies and guiding their decisions. For an overview of machine learning in the context of radiology, check out this source, but try not to get too caught up in the minute details (2).
Where Does Machine Learning Come In?
One example of the use of machine learning in radiology is in the early detection of disease. In a study published in Computational Intelligence and Neuroscience, the authors showed how effectively deep learning can detect breast cancer from mammograms (4). Specifically, they employed a ResNet-50 convolutional neural network to achieve 93% classification accuracy on the INbreast dataset. The authors report that this and similar approaches can serve as potentially lifesaving tools in the early diagnosis and classification of breast cancer, as evidenced by their high accuracy, which is unlikely to be replicated by any single human radiologist.
Perhaps the most obvious application of machine learning to radiology is for image interpretation. However, “lower hanging fruits,” such as determining the order of patients for scanning and scheduling, can see great benefit from this technology (5).
While machine learning might most obviously be applied to radiology in terms of image classification, “lower hanging fruits,” such as administrative tasks and operations management (e.g., determining patient scanning orders), could additionally benefit from this technology. Machine learning algorithms are capable of automating such processes, which would alleviate both staff and the healthcare system as a whole from these menial, time-consuming tasks (6). In this case, more emphasis could be placed on more complex duties, allowing for greater performance and efficiency across the board.
Looking Forward
As we ponder the impact of machine learning in radiology, we must be aware of the ethical considerations raised by this innovation. Data security and patient privacy are not to be taken lightly, and have been notorious concerns from the greater public regarding this technology. Moving forward, we must ensure that the implementation of radiological machine learning be accompanied by transparency and accountability. A great read on this topic is Ethics of AI in Radiology: European and North American Multisociety Statement (8) – we strongly encourage you to check it out.
Wrapping Up
Pairing machine learning and radiology has the potential to revolutionize many aspects of healthcare delivery, allowing us to unveil novel insights from medical images, improve diagnostic accuracy, and save more lives. Of course, this must be done with great caution, and we must see to it that these tools are deployed responsibly to sincerely benefit society.
References
Commentaires