AI in medical imaging: What Radiologist Should Know

The post-pandemic world has witnessed a surge in artificial intelligence (AI) transforming industries, and healthcare is no exception. With the potential to revolutionize medical diagnosis and patient care, AI has captured the imagination of clinicians and researchers alike. To delve into the practical realities of this exciting shift, Perceptra Company Limited hosted a timely seminar in July 2023: “From Theory to Practice: Sharing Practical Insights on AI in Clinical Radiology.” ‘AI in medical imaging: What Radiologist Should Know.’ led by Arunnit Boonrod, Doctor of Medicine (M.D.) and lecturer at the Department of Radiology, Khon Kaen University. Her expertise in both clinical practice and research fields bridged the gap between theoretical AI and its practical applications in radiology. With AI making waves in medicine, we’re excited to share this summary of the seminar on AI in medical imaging! Hope you find it interesting.

The basic understanding of AI

AI refers to the ability of machines to perform tasks that typically require human intelligence. There are various types of artificial intelligence that evolved over time from traditional programming to machine learning and to deep learning.
Programming VS Machine Learning [source]
In traditional programming, it refers to the conventional approach of manually writing code to create specific instructions for a computer to follow. So, there is no room for the system to learn independently. The traditional programming works best on problems with clear and deterministic set of logic.  Unlike traditional programing, machine learning  enables computers to learn from data with the desired output i.e. annotated data and make predictions without being explicitly programmed. The fundamental elements of machine learning include data, ground truth, and labeling, all contributing to the model’s accuracy. Deep learning is a subset of machine learning in which it divides its learning into small multiple tasks so called “layers.” Once it processes past the first layer, it transfers the learning to the next layer. Accumulating learning from millions of layers, it could perform some complex tasks. Inspired by how the human brain works, the basic concept of neural network aligns with that of our neurons. Neurons send signals to one another to transmit information to the next neuron to eventually drive to a desired outcome. Similarly a neural network transmit learning to the next. Usually neural network does not work in a single node, they typically aggregate into a layer.
The biological neuron graph & on the right: the artificial neural network [source]
Yet, in a neural network, the connections between neurons and the activation function hold vital significance in deciding whether information from one neuron will be transmitted to the next.

Trend and Current Application

THE BOOM AND BUST CYCLE OF AI RESEARCH [source]
The graph above shows the popularity of AI quite fluctuated. There are 2 AI winters when interest and funding of AI decreases. The field has experienced several hype cycles, disappointment and criticism, funding cuts, and renewed interest years or even decades later. “Hype is not absurd. Without hype, there wouldn’t be the introduction of new technology,” expressed Arunit, MD. In 2010, the ImageNet Challenge took place, marking an annual competition within the realm of computer vision. The challenge aimed to evaluate the performance of computer vision algorithms on tasks in tasks like large-scale object detection and image classification.
Error rate in the ImageNet Large Scale Visual Recognition Challenge [source]
The error rate in the ImageNet Large Scale Visual Recognition indicates a notable reduction in the error rate of AI over time, reaching a point where it surpassed human performance in 2015.

Current Application

The role of AI in the field of medicine is diverse and smoothly incorporated into our daily operations, including its application in modality operations to decrease acquisition time and radiation exposure, as well as in patient scheduling. Radiologists may be acquainted with the following AI tasks:
  1. Classification & Prediction
    1. Normal/Abnormal
    2. Benign / Malignant
    3. Geonomics
    4. Prognosis
  2. Object Detection
    1. Label lesion with boxes
  3. Segmentation
    1. Radiomics
    2. Quantification of thing
The task of segmentation may not be exciting, but it is quite useful, especially in the quantification of things. This helps avoid disagreements with colleagues about factors such as density and volume. For example, the AI from Perceptra– Inspectra CXR with CT ratio. This provides more meaningful information than simply stating ‘cardiomegaly.’

Threat or opportunity

‘Is AI a threat or opportunity?’ this is one of the most frequently asked questions. Arunit, MD. encourages to adopt a positive outlook on the inevitable technology like AI. This will not only applicable to radiologists or physicians, but it will impact across all industries. “AI is coming. We simply need to go with the flow.”  encouraged Arunnit, MD. She also reminded everyone that before AI, radiologists faced disruption from digital X-rays. During that time, some radiologists were against the change from analog to digital, while others embrace it. Eventually, it happened. “In any disruption, there are bound to be changes. However, from my point of view as a radiologist, as long as we’re experts in the field, I don’t think it will significantly impact us in our lifetime”

As she said “AI is coming. We need to go with the flow.”

What flow will you choose?

Flow    
Facebook
Twitter
LinkedIn

Related Resources

We believe a 'good AI' goes beyond just algorithms. Here's the characteristics of good AI from our perspective.
Perceptra today announced that our AI for chest X-ray analysis, Inspectra CXR has received official approval from the Health Sciences Authority (HSA) of Singapore.
The summary of Navigating the Waves of AI Technology in Healthcare led by Dr. Warasinee, a Faculty member, Institute of Field Robotics, King Mongkut's University of Technology Thonburi, and co-founder of Perceptra.