A framework for falsifiable explanations of machine learning models with an application in computational pathology

Schuhmacher, David; Schörner, Stephanie; Küpper, Claus; Großerueschkamp, Frederik; Sternemann, Carlo; Lugnier, Céline; Kraeft, Anna-Lena; Tannapfel, Andrea; Reinacher‐Schick, Anke; Gerwert, Klaus; Mosig, Axel

doi:10.1016/j.media.2022.102594

Cited by 8 publications

(20 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A Comparative Segmentation Network (CompSegNet) 5 was trained for each of the five dimensionality reduction methods. The network was trained on the full colon dataset described in section 2.3.…”

Section: Cancer Segmentation With Compsegnetmentioning

confidence: 99%

“…No training was required in the Random approach, in which the dimensions of the dataset were reduced to 16 randomly selected wavenumbers. All encoding approaches were compared to a neural network using full spectra (Schuhmacher et al, 2022). 5 In order to visualize and compare the latent space representations and decompositions, the t-SNE algorithm was applied to one test sample to further reduce the dimensions to Euclidean space.…”

Section: Encoding Comparison Using T-snementioning

confidence: 99%

“…1 Many studies have shown its analytical potential, par-ticularly in cancer research using either Fourier transform infrared (FTIR) microscopy 2,3 or, more recently, much faster quantum cascade laser (QCL) based infrared microscopy. 4,5 In an infrared microscopic image, the absorbance spectra at each pixel position represent fingerprints of the biochemical composition. As has been demonstrated in numerous studies, 3,[6][7][8] the highly resolved molecular information can be utilised to precisely localize different tissue components, 3 recognize tumor, 5 or distinguish different tumor subtypes 6,8 with high sensitivity and specificity.…”

Section: Introductionmentioning

confidence: 99%

“…4,5 In an infrared microscopic image, the absorbance spectra at each pixel position represent fingerprints of the biochemical composition. As has been demonstrated in numerous studies, 3,[6][7][8] the highly resolved molecular information can be utilised to precisely localize different tissue components, 3 recognize tumor, 5 or distinguish different tumor subtypes 6,8 with high sensitivity and specificity.…”

Section: Introductionmentioning

confidence: 99%

“…This difficulty motivates our comparative computational study in the present manuscript: we assess how different dimensionality reduction methods disentangle spectra into a limited number of components by measuring how the reduction affects subsequent classification. Since recent progress in computational pathology has been boosted significantly by deep convolutional neural networks, both for infrared imaging 5,11,12 and conventional histopathology imagery, 13 we focus on studying the effect of dimensionality reduction on such deep neural networks.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Dimensionality reduction for deep learning in infrared microscopy: a comparative computational survey

Müller,

Schuhmacher,

Schörner

et al. 2023

Analyst

View full text Add to dashboard Cite

show abstract

Section: Cancer Segmentation With Compsegnetmentioning

confidence: 99%

Section: Encoding Comparison Using T-snementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Dimensionality reduction for deep learning in infrared microscopy: a comparative computational survey

Müller,

Schuhmacher,

Schörner

et al. 2023

Analyst

View full text Add to dashboard Cite

show abstract

Causality and scientific explanation of artificial intelligence systems in biomedicine

Boge,

Mosig

2024

Pflugers Arch - Eur J Physiol

View full text Add to dashboard Cite

With rapid advances of deep neural networks over the past decade, artificial intelligence (AI) systems are now commonplace in many applications in biomedicine. These systems often achieve high predictive accuracy in clinical studies, and increasingly in clinical practice. Yet, despite their commonly high predictive accuracy, the trustworthiness of AI systems needs to be questioned when it comes to decision-making that affects the well-being of patients or the fairness towards patients or other stakeholders affected by AI-based decisions. To address this, the field of explainable artificial intelligence, or XAI for short, has emerged, seeking to provide means by which AI-based decisions can be explained to experts, users, or other stakeholders. While it is commonly claimed that explanations of artificial intelligence (AI) establish the trustworthiness of AI-based decisions, it remains unclear what traits of explanations cause them to foster trustworthiness. Building on historical cases of scientific explanation in medicine, we here propagate our perspective that, in order to foster trustworthiness, explanations in biomedical AI should meet the criteria of being scientific explanations. To further undermine our approach, we discuss its relation to the concepts of causality and randomized intervention. In our perspective, we combine aspects from the three disciplines of biomedicine, machine learning, and philosophy. From this interdisciplinary angle, we shed light on how the explanation and trustworthiness of artificial intelligence relate to the concepts of causality and robustness. To connect our perspective with AI research practice, we review recent cases of AI-based studies in pathology and, finally, provide guidelines on how to connect AI in biomedicine with scientific explanation.

show abstract

Local-to-global spatial learning for whole-slide image representation and classification

et al. 2023

Computerized Medical Imaging and Graphics

View full text Add to dashboard Cite

A framework for falsifiable explanations of machine learning models with an application in computational pathology

Cited by 8 publications

References 43 publications

Dimensionality reduction for deep learning in infrared microscopy: a comparative computational survey

Dimensionality reduction for deep learning in infrared microscopy: a comparative computational survey

Causality and scientific explanation of artificial intelligence systems in biomedicine

Local-to-global spatial learning for whole-slide image representation and classification

Contact Info

Product

Resources

About