Automatic discrimination of neoplastic epithelium and stromal response in breast carcinoma

Explainability and uncertainty: Two sides of the same coin for enhancing the interpretability of deep learning models in healthcare

BACKGROUND

The increasing use of Deep Learning (DL) in healthcare has highlighted the critical need for improved transparency and interpretability. While Explainable Artificial Intelligence (XAI) methods provide insights into model predictions, reliability cannot be guaranteed by simply relying on explanations.

OBJECTIVES

This position paper proposes the integration of Uncertainty Quantification (UQ) with XAI methods to improve model reliability and trustworthiness in healthcare applications.

METHODS

We examine state-of-the-art XAI and UQ techniques, discuss implementation challenges, and suggest solutions to combine UQ with XAI methods. We propose a framework for estimating both aleatoric and epistemic uncertainty in the XAI context, providing illustrative examples of their potential application.

RESULTS

Our analysis indicates that integrating UQ with XAI could significantly enhance the reliability of DL models in practice. This approach has the potential to reduce interpretation biases and over-reliance, leading to more cautious and conscious use of AI in healthcare.

Biological insights and novel biomarker discovery through deep learning approaches in breast cancer histopathology

Breast cancer remains a highly prevalent disease with considerable inter- and intra-tumoral heterogeneity complicating prognostication and treatment decisions. The utilization and depth of genomic, transcriptomic and proteomic data for cancer has exploded over recent times and the addition of spatial context to this information, by understanding the correlating morphologic and spatial patterns of cells in tissue samples, has created an exciting frontier of research, histo-genomics. At the same time, deep learning (DL), a class of machine learning algorithms employing artificial neural networks, has rapidly progressed in the last decade with a confluence of technical developments - including the advent of modern graphic processing units (GPU), allowing efficient implementation of increasingly complex architectures at scale; advances in the theoretical and practical design of network architectures; and access to larger datasets for training - all leading to sweeping advances in image classification and object detection. In this review, we examine recent developments in the application of DL in breast cancer histology with particular emphasis of those producing biologic insights or novel biomarkers, spanning the extraction of genomic information to the use of stroma to predict cancer recurrence, with the aim of suggesting avenues for further advancing this exciting field.

Integration of Deep Learning and Active Shape Models for More Accurate Prostate Segmentation in 3D MR Images

Magnetic resonance imaging (MRI) has a growing role in the clinical workup of prostate cancer. However, manual three-dimensional (3D) segmentation of the prostate is a laborious and time-consuming task. In this scenario, the use of automated algorithms for prostate segmentation allows us to bypass the huge workload of physicians. In this work, we propose a fully automated hybrid approach for prostate gland segmentation in MR images using an initial segmentation of prostate volumes using a custom-made 3D deep network (VNet-T2), followed by refinement using an Active Shape Model (ASM). While the deep network focuses on three-dimensional spatial coherence of the shape, the ASM relies on local image information and this joint effort allows for improved segmentation of the organ contours. Our method is developed and tested on a dataset composed of T2-weighted (T2w) MRI prostatic volumes of 60 male patients. In the test set, the proposed method shows excellent segmentation performance, achieving a mean dice score and Hausdorff distance of 0.851 and 7.55 mm, respectively. In the future, this algorithm could serve as an enabling technology for the development of computer-aided systems for prostate cancer characterization in MR imaging.

A content-based image retrieval system for the diagnosis of lymphoma using blood micrographs: An incorporation of deep learning with a traditional learning approach

Lymphomas, or cancers of the lymphatic system, account for around half of all blood cancers diagnosed each year. Lymphoma is a condition that is difficult to diagnose, and accurate diagnosis is critical for effective treatment. Manual microscopic analysis of blood cells requires the involvement of medical experts, whose precision is dependent on their abilities, and it takes time. This paper describes a content-based image retrieval system that uses deep learning-based feature extraction and a traditional learning method for feature reduction to retrieve similar images from a database to aid early/initial lymphoma diagnosis. The proposed algorithm employs a pre-trained network called ResNet-101 to extract image features required to distinguish four types of cells: lymphoma cells, blasts, lymphocytes, and other cells. The issue of class imbalance is resolved by over-sampling the training data followed by data augmentation. Deep learning features are extracted using the activations of the feature layer in the pre-trained net, then dimensionality reduction techniques are used to select discriminant features for the image retrieval system. Euclidean distance is used as the similarity measure to retrieve similar images from the database. The experimentation uses a microscopic blood image dataset with 1673 leukocytes of the categories blasts, lymphoma, lymphocytes, and other cells. The proposed algorithm achieves 98.74% precision in lymphoma cell classification and 99.22% precision @10 for lymphoma cell image retrieval. Experimental findings confirm our approach's practicability and effectiveness. Extended studies endorse the idea of using the prescribed system in actual medical applications, helping doctors diagnose lymphoma, dramatically reducing human resource requirements.

Breast histopathological image analysis using image processing techniques for diagnostic puposes: A methodological review

Breast cancer in women is the second most common cancer worldwide. Early detection of breast cancer can reduce the risk of human life. Non-invasive techniques such as mammograms and ultrasound imaging are popularly used to detect the tumour. However, histopathological analysis is necessary to determine the malignancy of the tumour as it analyses the image at the cellular level. Manual analysis of these slides is time consuming, tedious, subjective and are susceptible to human errors. Also, at times the interpretation of these images are inconsistent between laboratories. Hence, a Computer-Aided Diagnostic system that can act as a decision support system is need of the hour. Moreover, recent developments in computational power and memory capacity led to the application of computer tools and medical image processing techniques to process and analyze breast cancer histopathological images. This review paper summarizes various traditional and deep learning based methods developed to analyze breast cancer histopathological images. Initially, the characteristics of breast cancer histopathological images are discussed. A detailed discussion on the various potential regions of interest is presented which is crucial for the development of Computer-Aided Diagnostic systems. We summarize the recent trends and choices made during the selection of medical image processing techniques. Finally, a detailed discussion on the various challenges involved in the analysis of BCHI is presented along with the future scope.

Artificial Intelligence-Based Segmentation of Residual Tumor in Histopathology of Pancreatic Cancer after Neoadjuvant Treatment

Simple Summary

The use of neoadjuvant therapy (NAT) in patients with pancreatic ductal adenocarcinoma (PDAC) is increasing. Objective quantification of the histopathological response to NAT may be used to guide adjuvant treatment and compare the efficacy of neoadjuvant regimens. However, current tumor response scoring (TRS) systems suffer from interobserver variability, originating from subjective definitions, the sometimes challenging histology, and response heterogeneity throughout the tumor bed. This study investigates if artificial intelligence-based segmentation of residual tumor burden in histopathology of PDAC after NAT may offer a more objective and reproducible TRS solution.

Abstract

Background: Histologic examination of resected pancreatic cancer after neoadjuvant therapy (NAT) is used to assess the effect of NAT and may guide the choice for adjuvant treatment. However, evaluating residual tumor burden in pancreatic cancer is challenging given tumor response heterogeneity and challenging histomorphology. Artificial intelligence techniques may offer a more reproducible approach. Methods: From 64 patients, one H&E-stained slide of resected pancreatic cancer after NAT was digitized. Three separate classes were manually outlined in each slide (i.e., tumor, normal ducts, and remaining epithelium). Corresponding segmentation masks and patches were generated and distributed over training, validation, and test sets. Modified U-nets with varying encoders were trained, and F1 scores were obtained to express segmentation accuracy. Results: The highest mean segmentation accuracy was obtained using modified U-nets with a DenseNet161 encoder. Tumor tissue was segmented with a high mean F1 score of 0.86, while the overall multiclass average F1 score was 0.82. Conclusions: This study shows that artificial intelligence-based assessment of residual tumor burden is feasible given the promising obtained F1 scores for tumor segmentation. This model could be developed into a tool for the objective evaluation of the response to NAT and may potentially guide the choice for adjuvant treatment.

Computer assisted recognition of breast cancer in biopsy images via fusion of nucleus-guided deep convolutional features

BACKGROUND AND OBJECTIVE

Breast cancer is a commonly detected cancer among women, resulting in a high number of cancer-related mortality. Biopsy performed by pathologists is the final confirmation procedure for breast cancer diagnosis. Computer-aided diagnosis systems can support the pathologist for better diagnosis and also in reducing subjective errors.

METHODS

In the automation of breast cancer analysis, feature extraction is a challenging task due to the structural diversity of the breast tissue images. Here, we propose a nucleus feature extraction methodology using a convolutional neural network (CNN), 'NucDeep', for automated breast cancer detection. Non-overlapping nuclei patches detected from the images enable the design of a low complexity CNN for feature extraction. A feature fusion approach with support vector machine classifier (FF + SVM) is used to classify breast tumor images based on the extracted CNN features. The feature fusion method transforms the local nuclei features into a compact image-level feature, thus improving the classifier performance. A patch class probability based decision scheme (NucDeep + SVM + PD) for image-level classification is also introduced in this work.

RESULTS

The proposed framework is evaluated on the publicly available BreaKHis dataset by conducting 5 random trials with 70-30 train-test data split, achieving average image level recognition rate of 96.66  ±  0.77%, 100% specificity and 96.21% sensitivity.

CONCLUSION

It was found that the proposed NucDeep + FF + SVM model outperforms several recent existing methods and reveals a comparable state of the art performance even with low training complexity. As an effective and inexpensive model, the classification of biopsy images for breast tumor diagnosis introduced in this research will thus help to develop a reliable support tool for pathologists.

Stain Color Adaptive Normalization (SCAN) algorithm: Separation and standardization of histological stains in digital pathology

BACKGROUND AND OBJECTIVE

The diagnosis of histopathological images is based on the visual analysis of tissue slices under a light microscope. However, the histological tissue appearance may assume different color intensities depending on the staining process, operator ability and scanner specifications. This stain variability affects the diagnosis of the pathologist and decreases the accuracy of computer-aided diagnosis systems. In this context, the stain normalization process has proved to be a powerful tool to cope with this issue, allowing to standardize the stain color appearance of a source image respect to a reference image.

METHODS

In this paper, novel fully automated stain separation and normalization approaches for hematoxylin and eosin stained histological slides are presented. The proposed algorithm, named SCAN (Stain Color Adaptive Normalization), is based on segmentation and clustering strategies for cellular structures detection. The SCAN algorithm is able to improve the contrast between histological tissue and background and preserve local structures without changing the color of the lumen and the background.

RESULTS

Both stain separation and normalization techniques were qualitatively and quantitively validated on a multi-tissue and multiscale dataset, with highly satisfactory results, outperforming the state-of-the-art approaches. SCAN was also tested on whole-slide images with high performances and low computational times.

CONCLUSIONS

The potential contribution of the proposed standardization approach is twofold: the improvement of visual diagnosis in digital histopathology and the development of powerful pre-processing strategies to automated classification techniques for cancer detection.

Fully automated quantitative assessment of hepatic steatosis in liver transplants

BACKGROUND

The presence of macro- and microvesicular steatosis is one of the major risk factors for liver transplantation. An accurate assessment of the steatosis percentage is crucial for determining liver graft transplantability, which is currently based on the pathologists' visual evaluations on liver histology specimens.

METHOD

The aim of this study was to develop and validate a fully automated algorithm, called HEPASS (HEPatic Adaptive Steatosis Segmentation), for both micro- and macro-steatosis detection in digital liver histological images. The proposed method employs a hybrid deep learning framework, combining the accuracy of an adaptive threshold with the semantic segmentation of a deep convolutional neural network. Starting from all white regions, the HEPASS algorithm was able to detect lipid droplets and classify them into micro- or macrosteatosis.

RESULTS

The proposed method was developed and tested on 385 hematoxylin and eosin (H&E) stained images coming from 77 liver donors. Automated results were compared with manual annotations and nine state-of-the-art techniques designed for steatosis segmentation. In the TEST set, the algorithm was characterized by 97.27% accuracy in steatosis quantification (average error 1.07%, maximum average error 5.62%) and outperformed all the compared methods.

CONCLUSIONS

To the best of our knowledge, the proposed algorithm is the first fully automated algorithm for the assessment of both micro- and macrosteatosis in H&E stained liver tissue images. Being very fast (average computational time 0.72 s), this algorithm paves the way for automated, quantitative and real-time liver graft assessments.

Celiac disease diagnosis from videocapsule endoscopy images with residual learning and deep feature extraction

BACKGROUND AND OBJECTIVE

Videocapsule endoscopy (VCE) is a relatively new technique for evaluating the presence of villous atrophy in celiac disease patients. The diagnostic analysis of video frames is currently time-consuming and tedious. Recently, computer-aided diagnosis (CAD) systems have become an attractive research area for diagnosing celiac disease. However, the images captured from VCE are susceptible to alterations in light illumination, rotation direction, and intestinal secretions. Moreover, textural features of the mucosal villi obtained by VCE are difficult to characterize and extract. This work aims to find a novel deep learning feature learning module to assist in the diagnosis of celiac disease.

METHODS

In this manuscript, we propose a novel deep learning recalibration module which shows significant gain in diagnosing celiac disease. In this recalibration module, the block-wise recalibration component is newly employed to capture the most salient feature in the local channel feature map. This learning module was embedded into ResNet50, Inception-v3 to diagnose celiac disease using a 10-time 10-fold cross-validation based upon analysis of VCE images. In addition, we employed model weights to extract feature points from training and test samples before the last fully connected layer, and then input to a support vector machine (SVM), k-nearest neighbor (KNN), and linear discriminant analysis (LDA) for differentiating celiac disease images from heathy controls.

RESULTS

Overall, the accuracy, sensitivity and specificity of the 10-time 10-fold cross-validation were 95.94%, 97.20% and 95.63%, respectively.

CONCLUSIONS

A novel deep learning recalibration module, with global response and local salient factors is proposed, and it has a high potential for utilizing deep learning networks to diagnose celiac disease using VCE images.