The use of artificial intelligence and machine learning systems have in healthcare increased drastically in recent years. Analytics and computer vision techniques gained accuracy and are able to discern increasingly complex patterns.
One of the most transformative and powerful areas of machine learning applications within the healthcare sector is medical imaging. Computer vision systems are able to analyze medical images generated by imaging devices such as X-ray and MRI machines. They can help in the diagnosis by comparing images of healthy patients with those of patients diagnosed with a certain disease.
Let’s take a look at some of the most powerful applications of machine learning in medical imaging.
Despite its simplicity, MNIST is one of the most popular datasets for machine learning applications. MNIST is a collection of 70,000 images of handwritten digits. It is frequently used as a sanity-check database as it is easy to use and allows to gain some quick insight regarding the performance of an image recognition algorithm.
When it comes to identifying diseases, several recent studies have indicated that machine learning algorithms perform equally well or even better than clinicians in identifying potential anomalies in medical images. One application for computer vision in medical imaging, is the identification of cardiovascular anomalies. A machine learning system that analyzes images of heart tissue, such as a chest x-ray, could be employed to allow for a fast diagnosis and detection of cardiovascular diseases. In this way, x-ray images may not need to be sent to a specialist to obtain an analysis, but can be processed immediately by the algorithm thus saving valuable time.
If a patient is complaining about chest pain, a chest x-ray can be taken to detect cardiomegaly, an indicators for heart disease. Machine learning algorithms could also be used in combination with chest x-rays to detect an enlarged left atrial of the heart. This could rule out certain pulmonary and cardiac issues, and thus allows doctors to better allocate their resources by narrowing down possible diagnoses early on.
Injuries to the musculoskeletal system must be treated immediately and accurately, as failure to do so can lead to chronic pain. Even with the support of x-rays or other image modalities, such injuries are difficult to diagnose, as the detection of dislocations, soft-tissue damage, and small fractures can be challenging. When examining a medical image, a machine learning system can more easily detect small fractures than a clinician can.
AI tools proved to be effective in the detection of small, subtle variations within an image that a human observer may overlook. Similarly, machine learning systems can support clinicians in the follow-up care after major surgeries such as hip joint replacements. If a joint replacement loosens itself, complications can arise that require invasive, thus potentially harmful and expensive surgeries. While detecting potential problems with a joint replacement can be challenging for the human eye, AI systems could be used to identify and detect patterns unnoticed by humans and subsequently to reduce the false-negative rate.
Advances in the diagnosis of neurological diseases can help to improve conditions for patients with degenerative neurological diseases such as amyotrophic lateral sclerosis (ALS). Currently there is no cure for ALS, but progress regarding the diagnosis can help scientists in their search for a cure. Medical imaging is needed to identify ALS and to distinguish it from other diseases such as primary lateral sclerosis (PLS). The radiologist needs to examine medical images and interpret lesions within them, to decide whether the lesions in an image are indicative of ALS or PLS, or if it is simply a coincidence that they appear in similar locations.
Manual segmentation to indicate the motor cortex is often used in the diagnosis of ALS/PLS but is a time consuming and difficult process. Computer vision systems can automate the segmentation of images while improving both speed and accuracy of a diagnosis based on such images. Moreover, computer vision algorithms can mark images for a manual review by neurologists and clinicians. In addition, machine learning algorithms can be used to generate reports based on images themselves, thus reducing the workload health providers have to face.
One of the most common applications of medical imaging procedures are routine cancer screenings. Preventive screening for breast cancer and colon cancer are standard procedures, but medical images of these body regions can be difficult to interpret. An early sign for breast cancer are microcalcification of the breast tissue. AI systems that use quantitative imaging features to improve the detection of microcalcification could potentially minimize the number of unnecessarily performed biopsies.
Regarding the screening for colorectal cancer, AI can be employed in such routine checks to detect polyps, as polyps are precursors to cancer development. Radiologists may miss the development of such polyps, but an AI system could not only improve the screening result’s accuracy but also reduce the time needed for an examination.
If a patient already has an established cancer, AI can be used in the detection of malignancies/tumors. Deep learning algorithms can generate tumor probability heat maps based on medical images.. Such algorithms consider features such as location, density, shape, and area of tissue masses and can thus improve the tracking of tumors and changes in those tumors over time. One example of such a software is a tool introduced by the Fraunhofer Institute for Medical Image Computing (MEVIS), which tracks tumors and their changes throughout various images.
Magnetic Resonance Imaging (MRI) is one of the most powerful and useful medical imaging modalities, because it allows for the quantification and visualization of blood flow in a non-invasive way. MRIs facilitate the evaluation of conditions such as cardiovascular pathologies. Computer vision algorithms can be used to analyze MRI data more efficiently, thus reducing analysis time and freeing up medical resources.
One example of computer vision being used to analyze MRI images is a tool produced by Arterys, a deep learning medical imaging company. Arterys’ tool allows for the diagnosis of cardiovascular disease and drastically reduces the amount of time that is needed for the assessments of MRI scans.
Image datasets must be annotated before they can be used in the training of a computer vision classifier. Image annotation is the process of adding metadata to images, using techniques such as semantic segmentation or bounding boxes. Annotations need to be of high quality, as poorly annotated data will have a negative impact on the classifiers performance and may lead to misclassifications. A professional image annotation team will be able to ensure that your image database is properly annotated and that your image classifier will perform optimally, whether you are working with medical images or with images from another area of computer vision application.