Statistical approach to fine needle aspiration diagnosis of breast masses.

A statistical algorithm was used for recursively partitioning a consecutive series of 37 benign and 69 malignant fine needle aspirates to produce a decision tree for diagnosing breast masses to achieve optimal separation between benign and malignant cytology.

Abstract

A statistical algorithm was used for recursively partitioning a consecutive series of 37 benign and 69 malignant fine needle aspirates to produce a decision tree for diagnosing breast masses. Optimal separation between benign and malignant cytology was accomplished by evaluating clump characteristics when clumps were present and evaluating cell integrity when clumps were absent. The 1.5% false-negative and 9.7% false-positive rates obtained through this scheme are better than those reported for most series.