Mitotic Figure Recognition: Agreement among Pathologists and Computerized Detector

Abstract

Abstract. Despite the prognostic importance of mitotic count as one of the components of the Bloom -Richardson grade [3], several studies [2, 9, 10] have found that pathologists' agreement on the mitotic grade is fairly modest. Collecting a set of more than 4,200 candidate mitotic figures, we evaluate pathologists' agreement on individual figures, and train a computerized system for mitosis detection, comparing its performance to the classifications of three pathologists. The system's and the pathologists' classifications are based on evaluation of digital micrographs of hematoxylin and eosin stained breast tissue. On figures where the majority of pathologists agree on a classification, we compare the performance of the trained system to that of the individual pathologists. We find that the level of agreement of the pathologists ranges from slight to moderate, with strong biases, and that the system performs competitively in rating the ground truth set. This study is a step towards automatic mitosis count to accelerate a pathologist's work and improve reproducibility.

A comparison of algorithms and humans for mitosis detection

“…Due to its importance, a number of automated detectors have been designed to solve the problem [5,7,8,10,11].…”

“…Mitosis detection therefore represents a challenging visual pattern recognition task. In evaluating (on a large, non-public dataset) the agreement among different pathologists faced with this task, Malon et al found that "the level of agreement of the pathologists ranges from slight to moderate", with "strong biases", which "suggest that different pathologists interpret grading guidelines differently" (cited from [8]). This finding is consistent to the observation of low agreement among different pathologists on the evaluation set of the MITOS dataset (Goerner et al, oral presentation at ICPR 2012, unpublished).…”

“…In order to select such objects as C0 samples, we adopt a candidate selection approach, which is an integral part of most mitosis detection algorithms [10,11,8,7]. Candidates are selected by a simple, weak detector with high sensitivity and low specificity, which we learned from few images in the training set, then applied on the whole dataset.…”

“…We study how top-performing mitosis detection algorithms [4,5,6,7,8] in the recent ICPR2012 mitosis detection contest [9] compare with the performance of humans which are new to the problem. We design an user test that places such humans in the same conditions as algorithms (i.e.…”

Quantitative analysis of nuclear shape in oral squamous cell carcinoma is useful for predicting the chemotherapeutic response

“…In addition, several systems succeeded in analyzing morphology of cell nuclei on hematoxylin-eosin (HE) stained specimens. Researchers compete for accuracy to count mitotic cells [16][17][18]. Cosatto et al quantitatively graded nuclear pleomorphism on histological images [19].…”