Journal of Emergency and Internal Medicine

Artificial Intelligence Comparing Computer Vision Learning Models in Detecting Lung Cancer using CT imaging - A Metanalysis Review

Joint Event on 8^th Edition of International Conference & Exhibition on Pain Management, Physiotherapy & Sports Medicine & 9^th Edition of International Conference on Internal Medicine & Patient Care
March 18-19, 2020 London, UK

Benson A Babu, Kaviya Sathyakumar, Michael Munoz

Saint John's Esiscopal Hospital, USA

ScientificTracks Abstracts: J Emerg Intern Med

Abstract

Introduction: Computer Vision (AI) is a rapidly emerging tool in diagnostic medical imaging. We compare different deep learning and classic learning methods in lung cancer classification accuracy, sensitivity, and specificity. Furthermore, we evaluate the outcomes of deep learning methods to analysis by expert radiologists. Study Design • Meta-analysis • 300 Cross-sectional studies extracted with standard search strategies • 150 articles selected using Inclusion/Exclusion Criteria • PROSPERO Registered • PICO Framework • Search Engines PubMed/MEDLINE • EMBASE (or Scopus) • Cochrane Library • Google Scholar • Web of Science • IEEEXplore • DBLP Primary analysis: Deep learning methods vs. Classical machine learning methods for lung nodule/tumor CT detection and diagnosis Deep learning methods vs. analysis by radiologist Subgroup Analysis: Comparing specific types of Deep learning classifiers (e.g., CNN, DBN, auto-encoders, etc.) and/ or specific types of Classical machine learning methods (e.g., SVM, LDA, etc.) Outcomes Sensitivity: how well the algorithm recognizes the type of nodule correctly Specificity: measures the ability of the algorithm to remove the false positives, and a high specificity value means a low rate of misdiagnosis. Accuracy: measures the proportion of data that correctly classified. Sensitivity-specificity ROC curve and Area under the curve (AUC): other indicators used to evaluate the performance of a classifier. Data Analysis Grade Approach rating the quality of individual studies Heterogeneity testing using Cochran’s Q and I^2 Summary of Estimates and Forest Plot performed Examination of Publication Bias Conclusion: Deep learning methods are just as accurate, or more than compared to classic machine leaning methods. Deep learning are just as accurate as compared with physician radiologist. Subgroup analysis comparing specific classic machine and deep learning methods.

Biography

Benson Babu Completed his education from the Cleveland Clinic Foundation, earned hospital quality process improvement physician executive master’s in business administration. Areas of Active research include healthcare predictive analytics, machine learning, deep learning, computer-vision automation. He worked in NewYork-Presbyterian Hospital, USA and he is interested in Internal Medicine. He has published many papers in the journals.