Heuristic Oncological Prognosis Evaluator (HOPE): Deep-Learning Framework to Detect Multiple Cancers

Abstract

Cancer is the common name used to categorize a collection of diseases. In the United States, there were an estimated 1.8 million new cancer cases and 600,000 cancer deaths in 2020. Though it has been proven that an early diagnosis can significantly reduce cancer mortality, cancer screening is inaccessible to much of the world’s population. Machine learning approaches are increasingly successful in image-based diagnosis, disease prognosis, and risk assessment. A literature search with the Google Scholar and PubMed databases from January 2020 to June 2021 determined that currently, no machine learning model (n=0/417) has an accuracy of 90% or higher in diagnosing multiple cancers. We propose our model HOPE, the Heuristic Oncological Prognosis Evaluator, a transfer learning diagnostic tool for the screening of patients with common cancers. By applying this approach to magnetic resonance (MRI) and digital whole slide pathology images, HOPE 2.0 demonstrates an overall accuracy of 95.52% in classifying brain, breast, colorectal, and lung cancer. HOPE 2.0 is a unique state-of-the-art model, as it possesses the ability to analyze multiple types of image data (radiology and pathology) and has an accuracy higher than existing models. HOPE 2.0 may ultimately aid in accelerating the diagnosis of multiple cancer types, resulting in improved clinical outcomes compared to previous research that focused on singular cancer diagnosis.