AI in Healthcare: Medical Imaging, Digital Pathology, and Genomics
Mike McNamara
The promise of artificial intelligence (AI) is greater in healthcare than in almost any other industry. From improving patient outcomes and care to expanding the reach of medical expertise and reducing costs, the potential benefits are huge.
However, AI efforts in healthcare to date have barely scratched the surface of what will eventually be possible. The healthcare industry remains behind other industries in AI adoption. This lag is largely due to data privacy, data specificity, budget limitations, and access issues.
Three prominent use cases for AI in healthcare are medical imaging, digital pathology, and genomics. The use of AI in these use cases has not only increased the speed and accuracy of diagnosis, it has also enabled earlier detection of important diseases such as breast cancer. Although these technologies are independent, they are often employed together as part of an extended diagnostic workflow: Medical imaging leads to a biopsy, and examination of the biopsy results by a pathologist leads to a genomic study, which is used to develop a treatment plan that is personalized to the patient’s genome or observed genetic markers.
Medical Imaging
As a discipline, medical imaging is under significant pressure to increase efficiency. Patient populations are aging and experiencing conditions that require more imaging, but the size of the radiology workforce is flat or shrinking. Many countries have a shortage of radiologists, especially in rural areas. In the third world, lack of radiology expertise is a widespread problem.Whereas AI efforts across the medical imaging workflow are important to overall progress, it is the use of AI for image analysis and diagnosis—computer-aided diagnosis—that has captured the most attention. Deep learning models are being developed for a wide range of conditions, promising to increase the speed and accuracy of analysis and enable earlier disease detection. High-profile areas under study include detection of lung nodules, brain cancer, multiple sclerosis, breast cancer, and prostate cancer.
Digital Pathology
The pathology workforce is facing a situation very similar to that in medical imaging: Demand for pathology services is growing faster than the number of pathologists. This means that pathology labs must become more efficient to handle more cases in less time.In traditional pathology, slides are prepared from patient tissue samples and then reviewed by a pathologist under a microscope at high magnification. This manual process can be error prone and time consuming, especially if the pathologist needs to consult with an outside expert.
Although computational pathology is similar in many ways to AI in medical imaging, there are some significant differences. In general, digital pathology is several years behind medical imaging in terms of AI maturity. In part, this may be because pathology has been relatively slow to digitize. In medical imaging, digitization offered a clear path to cost reduction and increased workflow efficiency. However, digital pathology adds digital technology on top of existing physical processes, so cost benefits are less clear.
Genomics
Data management is a much more significant challenge in genomics than in medical imaging or digital pathology. With sequencing results for a single individual ranging up to 1TB in size, whole genome sequencing creates data management challenges in both research and clinical settings. Although the file formats used in genomics are standardized, there is no equivalent to a picture archiving and communication system (PACS) or veAIndor neutral archive (VNA) for managing sequence data.
This white paper examines in detail the challenges in each of these use cases, discusses the critical role that data plays, and describes possible approaches to address computing and data storage requirement
Mike McNamara
次のステップ
.png?width=5760&format=avif&disable=upscale)