AI in Medical Imaging for Beginners: I. Imaging Basics
TL; DR: Before covering hot topics like advanced AI algorithms applied to medicine, we need to do some ground work by covering the basics: how does imaging our body work and what are the basics modalities out there.
Disclaimer
This is my first Medium article ever, and therefore probably a short one, aiming to just provide a glimpse of what is possible in the medical world and aimed at tech enthusiasts like me that love the real-world applications that tech like AI can have in our lives and healthcare. As I have been working mainly in neuro-oncology this past year, you will find a few references to this area, although imaging is obviously broadly applicable.
Intro to Medical Imaging
Medical imaging plays an important role in diagnosing, staging, and monitoring our body, having been established as one of the pillars in modern healthcare, as it allows to visualize the internal structures of the body non-invasively. The most common imaging techniques include X-rays, computed tomography (CT) scans, magnetic resonance imaging (MRI), and ultrasound (US), with each modality providing unique and usually complementary insights.
Basic Principles
- X-rays: use ionizing radiation to create images of the body’s internal structures. X-rays are particularly effective for visualizing bones and detecting fractures. The traditional visualization is a 2D image.
- CT Scans: combine X-ray technology with computer processing to create detailed, 3D, and cross-sectional (frontal,sagittal, and axial) images of the body. It is considered the first-line imaging modality for patients presenting with neurological symptoms due to its widespread availability and rapid acquisition time. CT is particularly useful for detecting acute hemorrhage (bleedings) and calcification, making it useful to assess complex injuries. As it relies on ionizing radiation, like X-rays it may be unusable in certain occasions.
- MRI: uses strong magnetic fields and radio waves to produce detailed images of soft tissues like organs. It is considered the gold standard for brain, spinal cord and joints imaging due to its excellent soft tissue contrast and ability to provide detailed anatomical information. Conventional MRI sequences, such as T1-weighted, T2-weighted, and fluid-attenuated inversion recovery (FLAIR), help characterize the different tissue information, by leveraging the characteristics of the magnetic fields on the body.
- Ultrasound: employs high-frequency sound waves to create images of soft tissues and organs. It is commonly used for prenatal imaging and assessing heart conditions.
Some of these modalities have also more advanced versions. For example advanced MRI techniques, such as perfusion-weighted imaging (PWI) or diffusion-weighted imaging (DWI) provide additional information on tumor vascularity, cellularity, and metabolic profile, aiding for example in tumor grading and treatment planning. Other techniques, such as Magnetic Resonance Spectroscopy (MRS) rely on measuring certain particles in the body. MRS is usually used to assess the concentration of various metabolites within the brain, such as choline (a marker of cell membrane turnover) and N-acetylaspartate (a marker of neuronal integrity), which are essential tumor classification. The Positron Emission Tomography (PET) is a functional imaging modality that uses radiolabeled tracers to visualize metabolic activity within the body (usually the brain). 18F-fluorodeoxyglucose (FDG) is the most commonly used tracer, which reflects glucose metabolism and could hint for example to tumor locations. Functional MRI (fMRI) measures the changes in blood oxygenation level-dependent (BOLD) signals during specific tasks, helping to map eloquent brain areas (e.g., language, motor) prior to surgical planning. Diffusion Tensor Imaging (DTI), on the other hand, visualizes white matter tracts and can help assess the relationship between the tumor and critical white matter pathways, guiding surgical approach and minimizing post-operative deficits.
AI’s role in Medical Imaging
AI has been on the hype train for the last two years and it holds enormous potential by automating and enhancing various aspects of image analysis. The AI-powered algorithms have demonstrated exceptional precision in detecting abnormalities like tumors and fractures, often matching or surpassing human capabilities. These algorithms also contribute to improved image quality by reducing noise and enhancing resolution, simplifying the interpretation of complex images. Furthermore, AI’s predictive modeling capabilities utilize imaging data to forecast patient outcomes, assisting clinicians in making informed treatment decisions. This will not substitute the final clinical decision-making process, but rather help in patient management by being complementary to a doctors assessment.
The benefits are numerous: accelerated image analysis for faster diagnoses, more efficient workflows compared to traditional methods, consistent results with minimized variability, and its ability to detect subtle patterns that facilitate early detection of medical conditions. But it does not come without challenges and limitations, as the quality and diversity of training datasets significantly impact the performance of the models, with inadequate or biased data potentially leading to inaccurate results. That is why fairness, privacy, and trustworthiness in AI are becoming more and more relevant, with AI system necessitating rigorous validation and monitoring. The integration of AI into existing medical imaging practices also still presents challenges, requiring effective collaboration among technologists, clinicians, and developers.
Conclusion
This is just a first introduction into medical imaging and the possibilities of AI in this domain. Despite the named hurdles, the future of AI in medical imaging appears to be promising, paving the path in personalized medicine by merging classical imaging with other disciplines such as genomics or pathology.
I hope you enjoyed reading this brief introduction! In the next chapters we will cover the detailed basics of MRIs and how we need to process the data to build up AI projects. Stay tuned!
References
[1]https://onestepdiagnostic.com/houston-medical-imaging/
[2] https://www.facebook.com/brainline/. Brain Imaging: What Are the Different Types?— brainline.org. https://www.brainline.org/slideshow/brain-imaging-what-are-different-types. [Accessed 13–06–2024].
[3] T. Dandino. MRS: A Special Tool in Radiology at Cincinnati Children’s — Radiating Hope— radiologyblog.cincinnatichildrens.org. https:radiologyblog.cincinnatichildrens.org/mrs/. [Accessed 13–06–2024].
[4] S. S. Mannam, C. D. Nwagwu, C. Sumner, B. D. Weinberg, and K. B. Hoang. “Perfusion-weighted imaging: The use of a novel perfusion scoring criteria to improve the assessment of brain tumor recurrence versus treatment effects”. en. In: Tomography 9.3 (May2023), pp. 1062–1070.
[5] https://www.boldbusiness.com/health/artificial-intelligence-radiology-evolving/
