Novel chromatin texture features for the classification of pap smears

Abstract

This paper presents a set of novel structural texture features for quantifying nuclear chromatin patterns in cells on a conventional Pap smear. The features are derived from an initial segmentation of the chromatin into bloblike texture primitives. The results of a comprehensive feature selection experiment, including the set of proposed structural texture features and a range of different cytology features drawn from the literature, show that two of the four top ranking features are structural texture features. They also show that a combination of structural and conventional features yields a classification performance of 0.954±0.019 (AUC±SE) for the discrimination of normal (NILM) and abnormal (LSIL and HSIL) slides. The results of a second classification experiment, using only normal-appearing cells from both normal and abnormal slides, demonstrates that a single structural texture feature measuring chromatin margination yields a classification performance of 0.815±0.019. Overall the results demonstrate the efficacy of the proposed structural approach and that it is possible to detect malignancy associated changes (MACs) in Papanicoloau stain.

A Structural Texture Approach for Characterising Malignancy Associated Changes in Pap Smears Based on Mean-Shift and the Watershed Transform

“…Step 3: Feature extraction Structural texture features can then be derived from a quantitative characterisation of the attributes and arrangement of these resulting blobs [3], [5]. Features that can be computed for a blob include morphometric features (characterising the shape, size, position and boundary) computed from its binary mask and densitometric features computed from its intensity or optical density histogram.…”

“…Two variants of the proposed approach were considered: L p,BAIF and L p,BAIFcon . The approach we previously described in [3] served as a benchmark. It is essentially the approach proposed herein but such that Step 1 is replaced by the computation of an extended-depth-of-field (EDF) [8] image from all of the images in the Z-stack.…”

“…The set of chromatin-segmentation-derived features considered in the evaluation are the 67 structural texture features we previously described in [3] plus two new features: the median and the interquartile range of the contrast dynamics [9] of the regional minima associated with each dark blob in a nucleus. Another 96 nucleus features (not derived from the chromatin segmentation) from the literature were also considered: 92 conventional features described in [3] plus four more Hessianbased features described in [10].…”

“…Another 96 nucleus features (not derived from the chromatin segmentation) from the literature were also considered: 92 conventional features described in [3] plus four more Hessianbased features described in [10].…”

“…However such features do not correlate well with the terms and adjectives used by cytopathologists to describe chromatin (the part of the nucleus that takes on stain) texture: heterogeneity, granularity, margination, condensation, compaction, clearing, clumping, clod-like, diffuse, blobs, particles, granules, and particles. We recently introduced a class of structural texture features for nuclear chromatin [3]. The features are derived from an initial segmentation of the chromatin into blob-like texture primitives.…”