Machine Learning based histology phenotyping to investigate the epidemiologic and genetic basis of adipocyte morphology and cardiometabolic traits

Abstract

Genetic studies have recently highlighted the importance of fat distribution, as well as overall adiposity, in the pathogenesis of obesity-associated diseases. Using a large study (n = 1,288) from 4 independent cohorts, we aimed to investigate the relationship between mean adipocyte area and obesity-related traits, and identify genetic factors associated with adipocyte cell size. To perform the first large-scale study of automatic adipocyte phenotyping using both histological and genetic data, we developed a deep learning-based method, the Adipocyte U-Net, to rapidly derive mean adipocyte area estimates from histology images. We validate our method using three state-of-the-art approaches; CellProfiler, Adiposoft and floating adipocytes fractions, all run blindly on two external cohorts. We observe high concordance between our method and the state-of-the-art approaches (Adipocyte U-net vs. CellProfiler: R 2 visceral = 0.94, P < 2.2 × 10 −16 , R 2 subcutaneous = 0.91, P < 2.2 × 10 −16), and faster run times (10,000 images: 6mins vs 3.5hrs). We applied the Adipocyte U-Net to 4 cohorts with histology, genetic, and phenotypic data (total N = 820). After meta-analysis, we found that mean adipocyte area positively correlated with body mass index (BMI) (P subq = 8.13 × 10 −69 , β subq = 0.45; P visc = 2.5 × 10 −55 , β visc = 0.

Phenotyping of Klf14 mouse white adipose tissue enabled by whole slide segmentation with deep neural networks

“…Healthy white adipocytes may present inter-depot heterogeneity in size, with SAT cells generally larger than VAT (Boumelhem et al, 2017;Fang et al, 2015;Glastonbury et al, 2020); adipocyte diameter is associated with BMI in both depots, but with obesity traits adjusted for BMI only in VAT (Honecker et al, 2021). There is also a sex effect, as females have significantly smaller adipocytes than males in VAT (Glastonbury et al, 2020).…”

“…There is also a sex effect, as females have significantly smaller adipocytes than males in VAT (Glastonbury et al, 2020). Females have larger adipocytes than males in SAT (Glastonbury et al, 2020;Small et al, 2018), but not significantly after adjusting for BMI, age and ancestry (Glastonbury et al, 2020). Obesity, or excessive WAT expansion occurs through two mechanisms: initial increase in cell size (hypertrophy) followed by increase in cell numbers (hyperplasia), the latter having poorer prognosis (Hausman et al, 2001).…”

“…DeepCell (Bannon et al, 2019;Van Valen et al, 2016) replaces predetermined filters (Arganda-Carreras et al, 2017) by deep convolutional neural networks (DCNNs) that learn optimal feature extraction for the target cell and microscopy modality, although its fully connected layers restrict input images to a pre-arranged size. To overcome this, Adipocyte U-Net (Glastonbury et al, 2020) uses a Fully Convolutional Network (FCN) (Long et al, 2015), so that images of variable size can be processed, up to the GPU memory limit. DCNNs successfully tackle stain variability and other colour variations using their generalisation capabilities, transfer learning, data augmentation (geometric transformations and colour) or a combination thereof.…”

“…Segmentation by pixel classification (typically as background, cell boundary or cell interior) is widely used (Glastonbury et al, 2020;Sirinukunwattana et al, 2020;Van Valen et al, 2016), but the results may need to be regularised, for example with an active contour (Van Valen et al, 2016). In addition, segmentation results tend to be worse where membranes touch, have less definition or are damaged.…”

“…In addition, tissue areas are typically too large at full resolution to process in random-access or GPU memory, and need to be tiled and the results stitched together. Uniform titling is commonly used (Coudray et al, 2018;Glastonbury et al, 2020;Muñoz-Aguirre et al, 2020) as it is simple to implement, but the tile overlap needed to avoid dropping cells on the tile seams produces redundant computations; in this work, we propose an adaptive tiling algorithm to reduce that burden. Furthermore, it is necessary to differentiate between the cells of interest (mature white adipocytes) and other WAT components and image artefacts.…”