Abnormal localization of immature precursors (ALIP) detection for early prediction of acute myelocytic leukemia (AML) relapse.

Acute myelocytic leukemia (AML) is a relapsing and deadly disease. Thus, it is important to early predict leukemia relapse. Recent studies have demonstrated strong correlations of relapse with abnormal localization of immature precursors (ALIP). However, there is no related research on automated detection of ALIP so far. To this end, we have proposed an ALIP detection method to investigate the relevance with AML relapse. Kernelized fuzzy C-means clustering is applied first to separate the foreground (with cells) and background (without cells). Image repairing is then used to wipe out noises to mark region of interest. Then, image partition is introduced to separate the overlapping cells. After that, a set of features are extracted for the classification. Thereafter, support vector machine is applied to classify precursors. At last, filtering operations are applied to obtain the binary-precursor detection results. Thirty-seven patients with AML are examined. The results show that ALIP is efficiently detected in a high sensitivity and positive predictive value by our proposed method. The investigation also demonstrates the strong correlations of AML relapse with ALIP.