Computational integration of renal histology and urinary proteomics using neural networks

Lucarelli, Nicholas; Yun, Donghwan; Han, Dohyun; Ginley, Brandon; Moon, Kyung Chul; Rosenberg, Avi; Tomaszewski, John E.; Han, Seung Sook; Sarder, Pinaki

doi:10.1117/12.2613500

Cited by 2 publications

(7 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Following a priori data acquisition, segmentation, and handcrafted feature extraction (Section 3), each detected glomerulus was represented by a set of handcrafted features x glom ∈ ℝ 1 × d denoting global sclerosis, texture, color, and morphology. In this study, we build upon our previous feature extraction pipeline [53], which generates d = 316 numerical features per glomerulus. We obtained a single WSI per patient, and since each WSI was associated with a variable number of observable glomeruli, each patient (i.e., each sample in our dataset) was represented by , where g i is the number of distinct glomeruli segmented from patient i ’s WSI, and N is the number of patients in our dataset.…”

Section: Methodsmentioning

confidence: 99%

“…The second stage involved the extraction of (1) expert-defined color, texture, and morphological features, (2) binary indication of sclerosis, and (3) the two-dimensional glomerulus centroid location from each segmented glomerulus. In this study, we utilize an existing dataset of 56 patient biopsies that we curated in prior work [53]. A more detailed description of data acquisition and processing are described in Section 3.…”

Section: Histological Feature Extractionmentioning

confidence: 99%

“…We utilized an existing cohort of 56 patients with type 2 diabetes characterized in greater detail in our prior work studying the integration of histology with urinary proteomics [53]. Study subjects were from Seoul National University Hospital.…”

Section: Data Acquisitionmentioning

confidence: 99%

“…In this study, we build upon our previous feature extraction pipeline [53], which generates 𝑑𝑑 = 316 numerical features per glomerulus. We obtained a single WSI per patient, and since each WSI was associated with a variable number of observable glomeruli, each patient (i.e., each sample in our dataset) was represented by 𝑥𝑥 𝑖𝑖 ∈ ℝ 𝑔𝑔 𝑖𝑖 ×𝑑𝑑 ∀ 𝑖𝑖 = 1, 2, .…”

Section: Histological Feature Extractionmentioning

confidence: 99%

“…Following the segmentation step, we extracted 316 digital image feature types (e.g., sclerosis, color, texture, morphology, containment, and inter/intrastructural distances) for each segmented glomerulus as detailed in prior work [53]. All features were independently standardized using z-normalization (using only values from the training set of each cross-validation iteration) to yield a per-feature distribution with a mean of zero and standard deviation of one.…”

Section: Data Processingmentioning

confidence: 99%

See 4 more Smart Citations

Spatially Aware Transformer Networks for Contextual Prediction of Diabetic Nephropathy Progression from Whole Slide Images

Shickel

Lucarelli

Rao

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Diabetic nephropathy (DN) in the context of type 2 diabetes is the leading cause of end-stage renal disease (ESRD) in the United States. DN is graded based on glomerular morphology and has a spatially heterogeneous presentation in kidney biopsies that complicates pathologists predictions of disease progression. Artificial intelligence and deep learning methods for pathology have shown promise for quantitative pathological evaluation and clinical trajectory estimation; but, they often fail to capture large-scale spatial anatomy and relationships found in whole slide images (WSIs). In this study, we present a transformer-based, multi-stage ESRD prediction framework built upon nonlinear dimensionality reduction, relative Euclidean pixel distance embeddings between every pair of observable glomeruli, and a corresponding spatial self-attention mechanism for a robust contextual representation. We developed a deep transformer network for encoding WSI and predicting future ESRD using a dataset of 56 kidney biopsy WSIs from DN patients at Seoul National University Hospital. Using a leave-one-out cross-validation scheme, our modified transformer framework outperformed RNNs, XGBoost, and logistic regression baseline models, and resulted in an area under the receiver operating characteristic curve (AUC) of 0.97 (95% CI: 0.90-1.00) for predicting two-year ESRD, compared with an AUC of 0.86 (95% CI: 0.66-0.99) without our relative distance embedding, and an AUC of 0.76 (95% CI: 0.59-0.92) without a denoising autoencoder module. While the variability and generalizability induced by smaller sample sizes are challenging, our distance-based embedding approach and overfitting mitigation techniques yielded results that suggest opportunities for future spatially aware WSI research using limited pathology datasets.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Histological Feature Extractionmentioning

confidence: 99%

Section: Data Acquisitionmentioning

confidence: 99%

Section: Histological Feature Extractionmentioning

confidence: 99%

Section: Data Processingmentioning

confidence: 99%

See 3 more Smart Citations

Spatially Aware Transformer Networks for Contextual Prediction of Diabetic Nephropathy Progression from Whole Slide Images

Shickel

Lucarelli

Rao

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

Spatially aware transformer networks for contextual prediction of diabetic nephropathy progression from whole slide images

Shickel

Lucarelli

Rao

et al. 2023

Medical Imaging 2023: Digital and Computational Pathology

View full text Add to dashboard Cite

Diabetic nephropathy (DN) in the context of type 2 diabetes is the leading cause of end-stage renal disease (ESRD) in the United States. DN is graded based on glomerular morphology and has a spatially heterogeneous presentation in kidney biopsies that complicates pathologists’ predictions of disease progression. Artificial intelligence and deep learning methods for pathology have shown promise for quantitative pathological evaluation and clinical trajectory estimation; but, they often fail to capture large-scale spatial anatomy and relationships found in whole slide images (WSIs). In this study, we present a transformer-based, multi-stage ESRD prediction framework built upon nonlinear dimensionality reduction, relative Euclidean pixel distance embeddings between every pair of observable glomeruli, and a corresponding spatial self-attention mechanism for a robust contextual representation. We developed a deep transformer network for encoding WSI and predicting future ESRD using a dataset of 56 kidney biopsy WSIs from DN patients at Seoul National University Hospital. Using a leave-one-out cross-validation scheme, our modified transformer framework outperformed RNNs, XGBoost, and logistic regression baseline models, and resulted in an area under the receiver operating characteristic curve (AUC) of 0.97 (95% CI: 0.90-1.00) for predicting two-year ESRD, compared with an AUC of 0.86 (95% CI: 0.66-0.99) without our relative distance embedding, and an AUC of 0.76 (95% CI: 0.59-0.92) without a denoising autoencoder module. While the variability and generalizability induced by smaller sample sizes are challenging, our distance-based embedding approach and overfitting mitigation techniques yielded results that suggest opportunities for future spatially aware WSI research using limited pathology datasets.

show abstract

Computational integration of renal histology and urinary proteomics using neural networks

Cited by 2 publications

References 22 publications

Spatially Aware Transformer Networks for Contextual Prediction of Diabetic Nephropathy Progression from Whole Slide Images

Spatially Aware Transformer Networks for Contextual Prediction of Diabetic Nephropathy Progression from Whole Slide Images

Spatially aware transformer networks for contextual prediction of diabetic nephropathy progression from whole slide images

Contact Info

Product

Resources

About