Believe the HiPe: Hierarchical perturbation for fast, robust, and model-agnostic saliency mapping

Cooper, Jessica; Arandjelović, Ognjen; Harrison, David J.

doi:10.1016/j.patcog.2022.108743

Cited by 17 publications

(5 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Another area in which TB-AI falls behind is its deficiency in explainable artificial intelligence (XAI) techniques. Despite receiving considerable attention in multiple fields, including in healthcare and medical research [76], DL algorithms have not been widely implemented in clinical practice [77]. This is primarily due to the need for the enhanced transparency and interpretability of ML models, particularly in critical applications such as disease diagnosis and treatment.…”

Section: Discussionmentioning

confidence: 99%

Automated Methods for Tuberculosis Detection/Diagnosis: A Literature Review

Zachariou,

Arandjelović,

Sloan

2023

BioMedInformatics

Self Cite

View full text Add to dashboard Cite

Tuberculosis (TB) is one of the leading infectious causes of death worldwide. The effective management and public health control of this disease depends on early detection and careful treatment monitoring. For many years, the microscopy-based analysis of sputum smears has been the most common method to detect and quantify Mycobacterium tuberculosis (Mtb) bacteria. Nonetheless, this form of analysis is a challenging procedure since sputum examination can only be reliably performed by trained personnel with rigorous quality control systems in place. Additionally, it is affected by subjective judgement. Furthermore, although fluorescence-based sample staining methods have made the procedure easier in recent years, the microscopic examination of sputum is a time-consuming operation. Over the past two decades, attempts have been made to automate this practice. Most approaches have focused on establishing an automated method of diagnosis, while others have centred on measuring the bacterial load or detecting and localising Mtb cells for further research on the phenotypic characteristics of their morphology. The literature has incorporated machine learning (ML) and computer vision approaches as part of the methodology to achieve these goals. In this review, we first gathered publicly available TB sputum smear microscopy image sets and analysed the disparities in these datasets. Thereafter, we analysed the most common evaluation metrics used to assess the efficacy of each method in its particular field. Finally, we generated comprehensive summaries of prior work on ML and deep learning (DL) methods for automated TB detection, including a review of their limitations.

show abstract

Section: Discussionmentioning

confidence: 99%

Automated Methods for Tuberculosis Detection/Diagnosis: A Literature Review

Zachariou,

Arandjelović,

Sloan

2023

BioMedInformatics

Self Cite

View full text Add to dashboard Cite

show abstract

“…In this section we employ Hierarchical Perturbation (HiPe) [ 24 ] and standard iterative perturbation [ 25 ] to understand how the model is able to identify CD3 expressing lymphocytes. These methods are widely used for deep learning interpretability as they offer intuitive visual interpretations of which regions in the input were more or less important in determining the model’s output.…”

Section: Discussionmentioning

confidence: 99%

Lymphocyte Classification from Hoechst Stained Slides with Deep Learning

Cooper

Arandjelović

et al. 2022

Cancers

Self Cite

View full text Add to dashboard Cite

Multiplex immunofluorescence and immunohistochemistry benefit patients by allowing cancer pathologists to identify proteins expressed on the surface of cells. This enables cell classification, better understanding of the tumour microenvironment, and more accurate diagnoses, prognoses, and tailored immunotherapy based on the immune status of individual patients. However, these techniques are expensive. They are time consuming processes which require complex staining and imaging techniques by expert technicians. Hoechst staining is far cheaper and easier to perform, but is not typically used as it binds to DNA rather than to the proteins targeted by immunofluorescence techniques. In this work we show that through the use of deep learning it is possible to identify an immune cell subtype without immunofluorescence. We train a deep convolutional neural network to identify cells expressing the T lymphocyte marker CD3 from Hoechst 33342 stained tissue only. CD3 expressing cells are often used in key prognostic metrics such as assessment of immune cell infiltration, and by identifying them without the need for costly immunofluorescence, we present a promising new approach to cheaper prediction and improvement of patient outcomes. We also show that by using deep learning interpretability techniques, we can gain insight into the previously unknown morphological features which make this possible.

show abstract

“…For this, we use Hierarchical Perturbation (HiPe), 16 a saliency-mapping method which is both model-agnostic and highly computationally efficient – this is necessary in order to mitigate the huge computational cost of pixel-level attribution on gigapixel input images. Given that our model is dual-stage, combining first feature extraction followed by a classification network, and that in this work we tested different model architectures, patch sizes, and other hyperparameters, model agnosticism was also key in this choice of saliency algorithm.…”

Section: Methodsmentioning

confidence: 99%

Weakly supervised learning and interpretability for endometrial whole slide image diagnosis

Mohammadi

Cooper

Arandelović

et al. 2022

Exp Biol Med (Maywood)

Self Cite

View full text Add to dashboard Cite

Fully supervised learning for whole slide image–based diagnostic tasks in histopathology is problematic due to the requirement for costly and time-consuming manual annotation by experts. Weakly supervised learning that utilizes only slide-level labels during training is becoming more widespread as it relieves this burden, but has not yet been applied to endometrial whole slide images, in iSyntax format. In this work, we apply a weakly supervised learning algorithm to a real-world dataset of this type for the first time, with over 85% validation accuracy and over 87% test accuracy. We then employ interpretability methods including attention heatmapping, feature visualization, and a novel end-to-end saliency-mapping approach to identify distinct morphologies learned by the model and build an understanding of its behavior. These interpretability methods, alongside consultation with expert pathologists, allow us to make comparisons between machine-learned knowledge and consensus in the field. This work contributes to the state of the art by demonstrating a robust practical application of weakly supervised learning on a real-world digital pathology dataset and shows the importance of fine-grained interpretability to support understanding and evaluation of model performance in this high-stakes use case.

show abstract

Believe the HiPe: Hierarchical perturbation for fast, robust, and model-agnostic saliency mapping

Cited by 17 publications

References 14 publications

Automated Methods for Tuberculosis Detection/Diagnosis: A Literature Review

Automated Methods for Tuberculosis Detection/Diagnosis: A Literature Review

Lymphocyte Classification from Hoechst Stained Slides with Deep Learning

Weakly supervised learning and interpretability for endometrial whole slide image diagnosis

Contact Info

Product

Resources

About