Artificial Intelligence in Nephrology – From Non-Invasive Detection to Predictive Dialysis
Kidney disease is a severe health condition that results in high morbidity and economic burden. According to a report by Lancet, since 1990, chronic kidney diseases have increased by 29.3%, and in 2017, the global burden reached around 9.1%, which is about 700 million cases. Kidney disease varies from person to person in terms of disease presentation, progression, and treatment response. Its treatment costs are among the most expensive medical treatments, approximately 100 billion USD a year in the US alone.
There is no doubt that machine learning and artificial intelligence (AI) are very much part of the entire healthcare landscape. AI is increasingly being used to establish a more accurate diagnosis and predict prognosis for kidney diseases, both acute and chronic. The data sources for these include electronic health records (EHR), intraoperative physiological signals, ultrasound images of the kidney, and digitized biopsy specimens.
The application of AI in nephrology is based on personalized medicine that focuses on precise phenotype and outcome prediction. IgA nephropathy, also known as Berger’s disease, is a kidney disease that occurs when an antibody called immunoglobulin A (IgA) builds up in your kidneys. Professor Chen and a team of the National Clinical Research Center of Kidney Diseases, China, developed a risk prediction algorithm using machine-learning for immunoglobulin A nephropathy (IgAN). This system’s advantage is that the supervised machine-learning method requires an input of only a few numbers of predefined models for IgAN.
Many AI-based startups have developed systems that can detect and predict kidney diseases. RenalytixAI is one such artificial intelligence-enabled in vitro diagnostic company for kidney disease. It deals with early kidney disease detection and organ rejection. The company’s AI platform KidneyIntelIX combines diverse data inputs, including validated blood-based biomarkers, inherited genetics, and personalized patient data from the electronic health record, or EHR, systems, to generate a unique patient risk score. This patient risk score enables the prediction of progressive kidney function decline in chronic kidney disease.
Medial EarlySign is an Israel-based startup that can predict the development of kidney dysfunction in diabetic patients within a one-year time-frame using artificial intelligence’s machine learning-principles. The algorithm identifies at-risk patients by examining electronic health records, lab tests, medication, and other diagnostic parameters. Google’s DeepMind, a London-based lab, has developed an AI system that can predict the onset of acute kidney injury about two days before a physician would usually detect it. A deep recurrent neural network model has been used recently to predict acute kidney injuries (AKI) of inpatients using EHR, blood levels of creatinine and potassium, age, vital signs of heart rate, and oxygen saturation. This would give physicians more time to intervene and manage efficiently.
AI startup pulseData has developed predictive models using machine learning models for clinical purposes in patients with chronic kidney diseases. The Program for Education in Advanced Kidney Disease (PEAK), instituted by Rogosin Institute, is one such AI-based multidisciplinary program care team that assists patients in making a smooth transition to renal replacement therapy.
A Singaporean group of scientists led by Professor Charumathi Sabanayagam from The Singapore Epidemiology of Eye Diseases used data from three multiethnic populations to develop and validate an algorithm based on deep learning to detect chronic kidney disease non-invasively from retinal photographs. Similarly, another research team lead by Professor Kuo from China Medical University, Taiwan, developed an algorithm for automatically assessing the estimated glomerular filtration rate (eGFR) and chronic kidney disease status using a deep learning approach. They integrated the ResNet model pre-trained on an ImageNet dataset in our neural network architecture to predict kidney function based on ultrasound images of kidney and serum creatinine concentrations.
Multiclass segmentation and histopathologic assessment of kidney disease were reported using automated analysis of transplant biopsy using CNN in sections stained by periodic acid–Schiff by Professor Hermsen and a team from Radboud University Medical Center, Netherlands. Usually, manual scoring or traditional image-processing techniques are used to quantify and classify kidney diseases, which is labor-intense and time-consuming. Moreover, the network’s results have been validated with components from the Banff classification system.
The promising Future
Like any other medical condition, patients with kidney diseases are also positively impacted by AI and ML technologies. It is essentially based on extracting details from the massive but minute and heterogeneous data from the EHR and other parameters analyzed by the machine for patterns and relationships. Thus, risk stratification, predictive analytics, and clinical decision support tools can be developed. By leveraging AI, all the patient variables are put together with better decision making.
Machine learning algorithms are also being used for predictive analytics of blood-based biomarkers to detect kidney disease and predict dialysis and transplant. Such a predictive model will help to classify patients based on treatment needs and specialist care. There is a usual failure rate of 20% when a kidney transplant is done. AI, along with various biomarkers, can help in predicting the outcome of kidney transplantation. The pattern recognition of AI can help identify and monitor tissue rejection and personalize appropriate immune-suppressant medicines.
This is just the beginning of what AI has to offer the nephrology community. Personalized drug/therapy recommendations and lifestyle changes based on AI’s findings can stall the disease progression and avoid extensive treatments. The combination of AI, big data from lab reports, ultrasound images, and patient’s case history will completely change the way kidney diseases are managed.