Automatic discovery of clinically interpretable imaging biomarkers for Mycobacterium tuberculosis supersusceptibility using deep learning

Tavolara, Thomas E.; Niazi, M. Khalid Khan; Ginese, Melanie; Piedra-Mora, César; Gatti, Daniel M.; Beamer, Gillian; Gürcan, Metin N.

doi:10.1016/j.ebiom.2020.103094

Cited by 26 publications

(47 citation statements)

References 46 publications

(76 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1 depicts a flowchart for the overall methodology. Mice and M.tb infection, diagnostic categories of mice, slide preparation, and digital imaging protocols are same to our previous studies [48] .…”

Section: Methodsmentioning

confidence: 99%

“…At 8-10 weeks old, female DO mice (RRID:SCR_016408) were infected with very low (~20 bacilli) dose of M.tb strain Erdman (ATCC strain designation 38501) using a CH Technologies nose-only system. Mice were randomized in cages prior to infection [ 45 , 48 ]. After infection, mice were monitored daily for health, weighed at least weekly and euthanized when IACUC-approved removal criteria were met or at the predetermined experimental time point.…”

Section: Methodsmentioning

confidence: 99%

“…Training and validation of our model utilized a set of DO mice experimentally infected with M.tb from which H&E-stained lung biopsies and gene expression data (microarray) were acquired. We demonstrate that a modified version of our previous attention-based MIL model [48] that regresses to gene expression values is sufficiently accurate to discriminate SS from nSS better than our previous method when validated on a gradient boosted tree (GBT) classifier trained on ground-truth gene expression data.…”

Section: Introductionmentioning

confidence: 94%

“…This 'black-box' nature of deep learning limits its applicability in research and medicine, as both scientists and clinicians must know how decisions are made before informing a biological mechanism or clinical decision [ 1 , 47 ]. Attention-based deep MIL [ 48 , 58 ] offers a solution to each of these problems in deep learning when processing WSIs. A background on MIL and details regarding its attention-based implementation can be found in the Supplementary Methods.…”

Section: Model Descriptionmentioning

confidence: 99%

“…Our past work has utilized image analysis and deep learning to automatically detect histological features of M.tb- infected lungs of DO mice, including granulomas, cell-poor caseous necrosis, lymphocytic cuffs, macrophage-rich regions, neutrophil-rich regions, normal lung tissue, and acid-fast stained M.tb [ 35 , 37 , [45] , [46] , [47] ]. Our most recent work utilizes attention-based multiple instance learning (MIL) to classify DO mice into as "supersusceptible" (SS) and "non-supersusceptible" (nSS) with high accuracy (91.50 ± 4.68%) using only slide-level labels [48] . The work was particularly notable for its interpretability by human pathologists – examination of the MIL model "attention-weights" revealed that the model was making diagnostic decisions using a form of cellular necrosis: karyorrhectic and pyknotic nuclear debris.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Deep learning predicts gene expression as an intermediate data modality to identify susceptibility patterns in Mycobacterium tuberculosis infected Diversity Outbred mice

et al. 2021

Self Cite

View full text Add to dashboard Cite

Background Machine learning sustains successful application to many diagnostic and prognostic problems in computational histopathology. Yet, few efforts have been made to model gene expression from histopathology. This study proposes a methodology which predicts selected gene expression values (microarray) from haematoxylin and eosin whole-slide images as an intermediate data modality to identify fulminant-like pulmonary tuberculosis ('supersusceptible') in an experimentally infected cohort of Diversity Outbred mice (n=77). Methods Gradient-boosted trees were utilized as a novel feature selector to identify gene transcripts predictive of fulminant-like pulmonary tuberculosis. A novel attention-based multiple instance learning model for regression was used to predict selected genes' expression from whole-slide images. Gene expression predictions were shown to be sufficiently replicated to identify supersusceptible mice using gradient-boosted trees trained on ground truth gene expression data. Findings The model was accurate, showing high positive correlations with ground truth gene expression on both cross-validation ( n = 77, 0.63 ≤ ρ ≤ 0.84) and external testing sets ( n = 33, 0.65 ≤ ρ ≤ 0.84). The sensitivity and specificity for gene expression predictions to identify supersusceptible mice ( n =77) were 0.88 and 0.95, respectively, and for an external set of mice (n=33) 0.88 and 0.93, respectively. Implications Our methodology maps histopathology to gene expression with sufficient accuracy to predict a clinical outcome. The proposed methodology exemplifies a computational template for gene expression panels, in which relatively inexpensive and widely available tissue histopathology may be mapped to specific genes' expression to serve as a diagnostic or prognostic tool. Funding National Institutes of Health and American Lung Association.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 94%

Section: Model Descriptionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Deep learning predicts gene expression as an intermediate data modality to identify susceptibility patterns in Mycobacterium tuberculosis infected Diversity Outbred mice

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

Interpretability of Neural Networks with Probability Density Functions

Pan

Pedrycz

Cui

et al. 2022

Advcd Theory and Sims

View full text Add to dashboard Cite

It is an interesting topic to interpret artificial neural networks (ANNs) by considering some change various approaches. This paper explores the relationship between the input and output units of the simplest ANN, a single layer perceptron for the binary classification problem, from the probability point of view. If the feature variables of datasets follow independent normal distribution and outputs are activated by sigmoid function or smooth Relu function, we advocate that the probability density function (pdf) of the output variable is an exponential family distribution. Furthermore, by introducing an intermediate variable, the pdf of the output variable can be written as a linear combination of three normal distributions with same spread but different centers. Based on these results, the probability of the predicted class label can be written as a standard normal cumulative distribution function (cdf). The originality of this paper comes with interesting theoretical results to provide ANNs with a new description of the relationship between input variables to output variables, which can enable ANNs to be understood from a new perspective. Extensive experiments based on one artificial synthesized dataset and ten real‐world benchmark datasets validate the reasonability of those results.

show abstract