Virtual Biopsies, Digital Organs, Arrhythmia Prediction.
Artificial Intelligence in Health Care Roundup #12
“Any A.I. smart enough to pass a Turing test is smart enough to know to fail it.”
-Ian McDonald, River of Gods
1) Virtual Biopsy for Lung Cancer
Medical images, like CT scans, are rich with information we can’t see. Radiomics, is an evolving medical field that aims to extract large amount of quantitative features from medical images using characterization algorithms. Whereas a radiologist may use a hundred words to describe a tumor on a CT scan, an algorithm uses thousands of labels to describe the same image. With more precise ways to quantify tumor growth rate and patterns, researcher can begin to pick up the subtle patterns that may correlate with tumor type. Not all tumors are created equal.
Knowing the type of tumor helps doctors decide on the medications and surgery to treat the cancer. Today, identifying a tumor is a complicated process. A doctor must take a sample from the tumor (a biopsy), which is then sent to a pathologist who processes the tissue and looks at it under a microscope and then after about 2–3 weeks provides a diagnosis. This process is susceptible to many levels of human error and can often end up with frustrating results like: “inconclusive tumor, biopsy specimen insufficient for diagnosis”.
Therefore, the power of machine learning can be used to identify the radiological patterns specific to certain tumors. This could replace the need for physical biopsies in the future and allow for “virtual” biopsies.
In this study researchers applied machine learning principles to large data sets of lung CT scans to identify and segment lung cancer lesions.
2) Modeling Human Organs Biomechanics
Human organs are complex 3D structures. Each has its own biomechanical properties dictated by the types of tissue that comprise the organ. A breast is mostly fat and glands therefore much more elastic and compressible than the dense yet vascular tissue of the liver. Accurate models of the biomechanical properties of human soft tissue is important in many clinical applications, such as, radiotherapy administration or surgery. The current approach to create models uses the finite element method (FEM) due to its high accuracy. However, FEM is computationally very costly, and hence, its application in real-time or even off-line with short delays are still challenges to overcome. These researchers propose a framework based on Machine Learning to learn FEM modeling, thus having a tool able to yield results that may be sufficiently fast for clinical applications.
Read more: Machine Learning for modeling the biomechanical behavior of human soft tissue
3) From the Bottom of My Heart, Predict Arrhythmia
Cardiovasular diseases (CVDs) are the number one cause of death in the world especially the main reason of death in developed countries. People with CVDs or at high risk of CVDs can be managed by early detection of the disease and appropriate treatments. Therefore, the detection of CVDs for prevention and treatment is a significant task in medical health domain. A significant sign of heart disease, in particular an irregular heart rhythm is a of premature ventricular contraction (PVC). The PVC, also known as ventricular premature beat (VPB) is an irregular heartbeat that initiates at the bottom of the heart. A single PVC beat is commonly detected from healthy people and usually skipped without any symptoms. However, continuous PVC beats occasionally become a ventricular tachycardia (VT) that can lead to death.
These researchers used deep neural networks is proposed for the classification PVCs from heart monitor ECG data. Using TensorFlow, which is an open source machine learning platform initially developed by Google. they achieved 99.41% accuracy and a sensitivity of 96.08% with total 80,836 ECG beats including normal and PVC from the MIT-BIH Arrhythmia Database.
Read more: Premature Ventricular Contraction Beat Detection with Deep Neural Networks
