Radiology Imaging (PET/MRI) and Deep Learning
Diagnosis is about identifying disease, for example, looking at a chest X-ray and determine whether it contains disease, or identifying the locations of tumors in a brain MRI. Medical Diagnosis is the process of determining which disease or condition explains the person’s symptoms, signs and medical results.
This could be achieved using Pathology and Radiology data. A few interesting work done using deep learning, multi-omics and multi-modal imaging are:
Poor access to healthcare and errors in differential diagnosis represents a significant challenge to global healthcare systems. In US alone, an estimated 5% of outpatients are misdiagnosed every year. For patients with serious medical conditions, an estimated 20% are misdiagnosed at the level of primary care, out of which 1/3rd of the cases results in serious patient harm.
If machine learning is to help overcome these challenges, it is important that we first understand how diagnosis is performed and clearly define the desired output of our algorithms. Existing approaches, like Bayesian model-based and Deep Learning approaches, have conflated diagnosis with associative inference. While the former involves determining the underlying cause of a patient’s symptoms, the latter involves learning correlations between patient data and disease occurrences, determining the most likely diseases in the population that the patient belongs to. While this approach is perhaps sufficient for simple causal scenarios involving single diseases, it places strong constraints on the accuracy of these algorithms when applied to differential diagnosis, where a clinician chooses from multiple competing disease hypotheses. …