Algorithmic Assistance
By working with each other – and with carefully designed algorithms – we can improve patient outcomes
Chaith Kondragunta | | Longer Read
sponsored by AIRA Matrix
AIRA Matrix is a company that builds applications for the life sciences industry. I say life sciences because we focus on two broad aspects of pathology – the pharmaceutical drug discovery space and the medical diagnostic space. For both, we develop deep learning algorithms that analyze pathology samples, automate work processes, and help increase laboratories’ workflow efficiency.
Higher accuracy is crucial in laboratory processes that aid critical decision-making – for instance, in discovery pathology or oncopathology. Our deep learning-based algorithms function as more objective and accurate quantitative tools in such applications. In other instances, such as in preclinical toxicology studies, labs may need to improve turnaround times. We have solutions that help reduce the resource requirement for such tasks – which currently take days – to minutes. With our solutions, pathologists have the advantage of increasing accuracy or speed and, quite often, both.
One final hurdle for labs: healthcare costs are growing and pathologists are under constant pressure to reduce costs. Not all tasks are cost-effective when performed by a human pathologist – and we feel artificial intelligence (AI) and automation can help in these instances.
Aiding the big decisions
In addition to solutions that improve the speed and accuracy of diagnosis, we have initiatives underway to aid development of novel prognostic and predictive markers. Our vision is solutions that help pathologists practice precision diagnostics to aid disease stratification, patient risk stratification, and treatment selection. To this end, we are developing predictive and prognostic algorithms in collaboration with a number of research partners and hospital systems. Our goal is to reach a point where we can help caregivers confidently conclude, “This is the patient’s personalized therapy recommendation based on their condition, the progress of the disease, and the outcomes seen in patients with similar characteristics.”
Consider prostate cancer; one of the first steps in grading the disease is Gleason scoring – which, despite its ubiquity, is not a perfectly applied system. When used by a single pathologist, its accuracy is sometimes lacking; when used by multiple pathologists, Gleason scoring is subject to interobserver variability. In response, we have developed tools that can help score the disease more objectively. Further, important prognostic parameters like tumor volume are currently eyeball assessments. To overcome this shortcoming, our tools offer accuracy and reproducibility, so that every pathologist who looks at the model will have the same objective information to assist their decision-making. Ultimately, this will lead us to the point where we can make predictions about the course of a patient’s disease and suggest a treatment approach based on a consensus between the pathologists – and the algorithms.
A platform-based service model
We recognize the complexity of these initiatives and have adopted a platform- based approach to increase the chances of success. Our approach allows us to develop custom solutions faster and more efficiently based on users’ needs. We analyze data, respond to the results of analyses, and receive feedback from our users that lets us further improve the platform. Customizability and responsiveness aren’t the only reasons we chose a platform- based strategy; another advantage is the ease of collaboration within and across disciplines. Caregivers and pathologists work on complex, often multisystem problems and deal with multimodal data – having a platform for collaboration helps break down silos and facilitates comprehensive solutions.
We are doing exciting work on improving patient outcomes in oncology – primarily in prostate and lung carcinoma. In lung carcinoma, we can analyze transbronchial aspirates and provide assessments in minutes. In current practice, oncologists and pathologists often work separately, need multiple aspirate-gathering attempts, and the overall turnaround time can be a few days! Our solution makes it possible for oncologists and pathologists to work together in real time as procedures are performed on patients. When the oncologist, pathologist, and algorithms are in sync, diagnosis and treatment selection happen faster – with better patient outcomes.
To make a meaningful difference in the diagnostic process, we have adopted the approach of taking what has been developed in conventional pathology – which is often the gold standard, but invasive – and make it multimodal by incorporating technologies such as radiology or genomics. This lets us create meaningful solutions that make a difference to patients. For this approach to be successful, we seek to partner with leading institutions worldwide. With these collaborations across laboratories, clinics, and industry, we aim to create diagnostic solutions that are less invasive, more accurate, and provide better outcomes than current practices.
