Premium
Medical pattern recognition pitfalls in a clinical setting: Renal cell carcinoma survival prediction
Author(s) -
KimmeSmith Carolyn,
Cochran Sachiko T.,
Roe Denise J.
Publication year - 1988
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.596265
Subject(s) - medicine , renal cell carcinoma , radiology , normalization (sociology) , survival analysis , nuclear medicine , oncology , sociology , anthropology
Three studies to predict renal cell carcinoma patient survival from tumor textures found in digitized early venous phase arteriograms were successful individually. However, when the three methods were compared, they were not consistent and no single method was clinically useful. The first study predicted 5 yr survival of 37 patients with 87% accuracy. The second study added 29 patients to the data base; the poor survival of the 21 patients who died within 5 yr of diagnosis was predicted with 80.9% accuracy. When 27 of these cases were redigitized with a laser scanner, average survival prediction accuracy was 78%. In these studies, digitization hardware, radiographic technique, normalization methods, window selection, and contrast medium distribution all contributed to differences in the statistics separating poor from good patient survival.