Autonomous Medicine: Computer-Assisted Diagnosis Cancer Detection Sharmila Anandasabapathy, MD Professor of Medicine in Gastroenterology Director, Baylor Global Health Baylor College of Medicine
Challenges in Endoscopic Cancer Screening in Remote Settings Globally Clinicians: less experienced Environment: infrastructure power geography: loss of follow up, no “back up” Cost Technical support: limited
Portable Microendoscopy: Experiences in China, Honduras, Africa
Image Analysis Software Blue outline – ROI Red outline – enlarged/abnormal nuclei 056_27_007 Yellow outline – nuclei
Normal 02_01_006-normal
Normal (with software) overlay % abnormal nuclei: 13.5
Early Cancer 164_80_001-high grade
Early Cancer (Software Overlay) % abnormal nuclei: 30.9
Summary No significant difference in pre- and post-software read among experts Significant increase in accuracy among novices Specificity went up significantly This leads to greatest improvement in “yield” & impact “False positives” are an issue with standard screening Increased confidence among all providers = more likely to act on diagnosis (treat, biopsy, NOT biopsy, etc.)
What does this Mean for Software-Assist Clinician Confidence & Impact A software-assisted diagnosis increases confidence uniformly Confidence translates to point-of-care decisions Experience of Provider is Key: Who is user? Novices rely on software Experienced providers rely on themselves Implications for Space and Remote Settings? Skill level is critical: Novices will rely on software (often regardless of accuracy of software) Human Factors element: Effect of stress on confidence and decisions