The Covid-19 pandemic has had a profound impact across industries and healthcare in particular – every aspect of it is changing- from diagnosis to treatment and through the entire continuum of care. This has also created an urgency in the healthcare industry to look for innovative solutions and boost the faster, efficient application of technologies like Artificial Intelligence (AI) and deep learning. Pathology is one area that stands to benefit from these applications.
Pathologists today spend a significant amount of time observing tissue samples under a microscope, and they are facing resource shortages, growing complexity of requests, and workflow inefficiencies with the growing burden of diseases. Their work underpins every aspect of patient care, from diagnostic testing and treatment advice to the use of cutting-edge genetic technologies. They also have to work together in a multidisciplinary team of doctors, scientists, and healthcare professionals to diagnose, treat, and prevent illness. With increasing emphasis on sub-specialization, specialists’ taking a second opinion means shipping several glass slides across laboratories, sometimes to another country. This means reduced efficiency and delayed diagnosis and treatment. The current situation has disrupted this workflow.
Digitization in pathology
Digitization in pathology has enabled an increase in efficiency, speed, and enhanced quality of diagnosis. Recent technological advances have accelerated the adoption of digitization in pathology, similar to the digital transformation that radiology departments have experienced over the last decade. Digital pathology has enabled the conversion of the traditional glass slide to a digital image, which can then be viewed on a monitor, annotated, archived, and shared digitally across the globe for consultation based on organ sub-specialization. A vast data set has become available with digitization, supporting new insights to pathologists, researchers, and pharmaceutical development teams.
The promise of AI
The availability of vast data enables the use of Artificial Intelligence methods to transform further the diagnosis and treatment of diseases at an unprecedented pace. Human intelligence assisted with artiﬁcial intelligence can provide a well-balanced view of what neither of them could do on their own. The evolution of deep learning neural networks and improved accuracy for image pattern recognition have been staggering in the last few years. Like how we learn from experience, the deep learning algorithm would perform a task repeatedly, each time improving it a little to achieve a more accurate outcome.
The approach to diagnosis that incorporates multiple sources of data (for example, pathology, radiology, clinical, molecular, and lab operations) and using mathematical models to generate diagnostic inferences and presenting with clinically actionable knowledge to customers is computational pathology. Computational pathology systems can correlate patterns across multiple inputs from the medical record, including genomics, enhancing a pathologists’ diagnostic capabilities to make a more precise diagnosis. This allows pathologists to eliminate tedious and time-consuming tasks while focusing more on interpreting data and detailing the implications for a patient’s diagnosis.
AI applications that can easily augment a pathologist’s cognitive ability and save time are, for example, identifying the sections of significant interest in biopsies, finding metastases in the lymph nodes of breast cancer patients, counting mitoses for cancer grading, or measuring tumors point-to-point. Going forward, the ultimate goal is to integrate all these tools and algorithms into the existing workflow and make it seamless and more efficient.
However, Artificial Intelligence in pathology is quite complex. The IT infrastructure required in terms of data storage, network bandwidth, and computing power is significantly higher than radiology. Digitization of Whole Slide Images (WSI) in pathology generate large amounts of gigapixel sized images, and processing them needs high-performance computing. Training a deep learning network on a whole slide image at full resolution can be very challenging. With the increase in the processing power with GPUs, there is a promise to train deep learning networks successfully, starting with training smaller regions of interest.
Another key aspect of training deep learning algorithms is the need for large amounts of labeled data. For supervised learning, a ground truth must first be included in the dataset to provide an appropriate diagnostic context, and this will be time-consuming. Obtaining adequately labeled data by experts is the key.
Digitization in pathology supported by appropriate IT infrastructure enables pathologists to work remotely without the need to wait for glass slides to be delivered and maintaining social distancing norms. The promise of Artificial Intelligence will only further accelerate the seamless integration of algorithms into the existing workflow. These unprecedented times have raised many challenges and provide us a chance to accelerate the application of AI and, in turn, achieve the quadruple aim — enhancing the patient experience, improving health outcomes, lowering the cost of care, and improving the work-life of care providers.