Machine Learning for Early Diagnosis of Rare Diseases: A Review

Abstract

Diseases with a lot in common but differ a lot in symptoms, age of onset, translation, and more are rare diseases. But they affect 3.5% to 5.9% of the population at some point in life. A lot of rare diseases can cause life-threatening or severely debilitating diseases without timely treatment, so the early diagnosis of these diseases is vital for better outcomes in patients. Most rare diseases take more than three months before being diagnosed. More than half of patients wait a year or more. Patients undergo a barrage of tests, many of which may be of little help due to high false-negative rates. Even when correctly diagnosed, the treatment pathways remain unknown. A major hurdle is the unawareness of general practitioners, who hence send patients to specialized centers. However, these centers can only take on a limited number of cases, delaying treatment. To increase accuracy and efficiency, diagnostic processes must be automated. Machine learning (ML) can help us model vast amounts of data to identify complex correlations. This paper outlines an approach based on machine learning that combines patient and test data with the help of an expert system to improve diagnostic decision-making with minimum delay and better patient satisfaction.