Nephrology Department

Within the healthcare landscape, the nephrology department stands as a bastion of expertise dedicated to the intricate realm of kidney health. Nephrologists, the specialized physicians in this domain, focus their efforts on comprehending the complexities of renal function, disorders, and therapies. With an unwavering commitment to safeguarding this vital organ, the nephrology department emerges as a cornerstone of patient care.


In the heart of nephrology lies nephropathology, a discipline that peels back the layers of kidney tissue to uncover the hidden stories within. Armed with cutting-edge techniques, nephropathologists meticulously examine microscopic and macroscopic characteristics. Through this exploration of kidney biopsies, they illuminate the enigmas of kidney diseases, unraveling their mechanisms and impacts.

Exploring Kidney Diseases and Advanced Imaging Techniques

In the intricate landscape of kidney health, certain diseases stand as focal points of research and clinical attention. Among these are ANCA-associated vasculitis, IgA nephropathy, glomerulonephritis, and tubulointerstitial nephritis. Each disease brings forth its unique challenges, demanding meticulous diagnosis and tailored therapeutic strategies.

ANCA-Associated Vasculitis

ANCA-associated vasculitis encompasses a group of autoimmune disorders affecting small blood vessels. These diseases, driven by antineutrophil cytoplasmic antibodies (ANCA), lead to inflammation and damage in various organs, including the kidneys. Accurate diagnosis is essential for prompt intervention to manage kidney involvement and mitigate systemic complications.

IgA Nephropathy

IgA nephropathy, also known as Berger’s disease, is a kidney disorder characterized by the accumulation of immunoglobulin A (IgA) in the glomeruli, the tiny filtering units within the kidneys. This buildup of IgA can lead to inflammation and damage in the kidneys, affecting their ability to filter waste and excess fluids from the blood.

A H&E staining (200 ×), showing red blood cell casts in some tubular lumens. B Light microscopy (PAS 200 ×), revealing a mesangioproliferative pattern of glomerulonephritis with massive mesangial and endocapillary hypercellularity. C Immunohistochemical positivity for IgA (red) in mesangial areas and at few peripheral glomerular basement membranes (400 ×). D Immunohistochemical positivity for C3 (red) in mesangial areas (400 ×)


Glomerulonephritis encompasses a range of conditions characterized by glomerular inflammation and damage. It can stem from various causes, including infections, autoimmune reactions, and genetic predispositions. The challenge lies in precise classification and identification of the underlying cause to tailor treatments effectively.


Advanced Imaging Techniques: CT and Ultrasound

While CT (computed tomography) and ultrasound are valuable techniques for diagnosing chronic kidney disease (CKD), our focus shifts to a more intricate dimension. Our research endeavors pivot towards the microscopic bright-field images that serve as windows into the cellular intricacies of CKD. Through these images, we unravel the cellular landscape, exploring the structural and functional changes that underlie CKD progression.

Microscopic Bright-Field Images: Illuminating CKD Insights

These images, acquired through advanced microscopy techniques, guide our clinical research. They offer unprecedented clarity into cellular alterations, aiding in the development of targeted treatments and therapies. By delving deep into the microscopic realm, we uncover the subtle transformations that bear direct relevance to patient care. This convergence of technology, research, and patient-centric focus propels us towards a future where CKD management is refined and patient outcomes are optimized.

This comprehensive approach underscores our commitment to advancing medical understanding, driving innovation, and ultimately enhancing the lives of patients grappling with CKD.

The advanced software, HSA KIT, equipped with the specialized module “Kidney Analyzer,” emerges as a powerful tool in the realm of medical image analysis. This module stands out for its profound focus on the precise segmentation of kidney compartments, such as glomerulus, tubulus, and vessels. Notably, it extends its capabilities to sub-segment these compartments into finer constituents.

The Glomerulus Matrix receives meticulous attention within the software, as it skillfully demarcates intricate components including Bowman’s space, capsule, capillaries, and cells. This analysis is conducted through various stains, with a particular emphasis on Hematoxylin and Immunohistochemistry (IHC) combinations.

The primary spotlight of this endeavor falls upon the Hematoxylin and IHC stain combination. This specific approach brings forth a wealth of insights, enabling the software to meticulously categorize kidney compartments. Notably, this categorization extends to diverse variations of glomeruli, tubules, and cells, encompassing pivotal entities like podocytes, epithelial cells, and mesangium cells.

This cutting-edge software embodies an exceptional blend of precision and versatility, catering to the intricate landscape of nephrology research. By harnessing the potential of HSA KIT’s Kidney Analyzer module, medical professionals and researchers can delve into the nuanced world of kidney histopathology. The tool not only demystifies the complexity of kidney compartments but also provides a comprehensive understanding of the variations present within these structures.

In essence, the software is a formidable asset that enables the exploration and analysis of kidney compartments with unparalleled depth and

accuracy. Its ability to navigate through the finer details of staining combinations, especially Hematoxylin with IHC, empowers researchers to uncover insights that were once elusive. This module’s prowess in differentiating and categorizing glomeruli types, tubules, and various cell components elevates the field of nephrology research.

The Kidney Analyzer module within HSA KIT exemplifies the fusion of cutting-edge technology with the intricacies of medical science. It serves as a guiding light for those seeking to unravel the mysteries of kidney structure, paving the way for advancements in diagnosis, treatment, and understanding kidney-related pathologies.

Data of tubules

Deciphering File Names: C3, C2, C4b, CFB, IgA, and OT

In our quest to unravel the intricacies of kidney health, file names like C3, C2, C4b, CFB, IgA, and OT hold pivotal significance. These abbreviations signify critical components within the context of our research, each carrying unique insights:

  • C3, C2, C4b: These abbreviations refer to complement system proteins involved in immune responses. C3, C2, and C4b are key players in the cascade of reactions that form the immune response, impacting various aspects of kidney health and function.
  • CFB: This refers to Complement Factor B, another essential component of the complement system. CFB plays a role in the alternative pathway of the complement cascade, contributing to immune responses and inflammation.
  • IgA: Immunoglobulin A (IgA) is an antibody that plays a critical role in immune defense, particularly at mucosal surfaces. In the context of kidney health, detecting IgA deposits in images can indicate IgA nephropathy, a kidney disease.


zoom out :Tubule

zoom in :Tubule

      Whole slide images (WSI)

Whole slide imaging (WSI) or virtual microscopy is a technology to scan and digitally archive slides in high resolution. Slide scanners take separate images of each field of view across the entire microscopy slide. The individual pictures are then stitched together to generate a single

Cancer Diagnosis and Cancer Staging can only be carried out with the help of the Whole Slide Images(WSIs). Therefore it is very important to know about WSI and its acquisition.

Digitized Tissue Slides are known as Whole Slide Image(WSI) these WSIs comprise of Multi Giga Pixels. The digitization process steps comprise of:

  • Extraction of Biopsy Sample from Patients Body
  • Placing Tissue Samples on Glass Slides
  • Staining of Glass Slides
  • Scanning of Glass Slides via Digital Slide Scanner

     Deep learning (DL)

Deep learning is a subset of machine learning which is essentially a neural network with three or more layers. These neural networks attempt to simulate the behavior of the human brain—albeit far from matching its ability—allowing it to “learn” from large amounts of data. While a neural network with a single layer can still make approximate predictions, additional hidden layers can help to optimize and refine for accuracy.

Deep learning drives many artificial intelligence (AI)  applications and services that improve automation, performing analytical and physical tasks without human intervention.

               Explainable DL (xDL)

Making complex deep learning models more transparent and understandable for humans is known as explainable deep learning (XDL). It involves methods that shed light on the processes these models use to make their judgments or predictions. By providing understandable justifications for the actions of the model, XDL seeks to improve trust, accountability, and the capacity to diagnose model behavior. This is particularly significant in crucial areas where developing trust and adhering to regulations depend on understanding the logic behind AI decisions.

  • Certification      

Certification is frequently essential in the field of medical and diagnostic technologies to make sure that the instruments, processes, or algorithms used for diagnosis or treatment adhere to certain standards of accuracy, efficacy, and ethical considerations. When it comes to deep learning models and their use in diagnosing medical conditions, certification may involve determining whether the model complies with specific performance standards, moral principles, and legal requirements.

Depending on the industry, application, and regulatory context, certification standards may change. Your thesis topic focuses on the application of certification standards to the use of explainable deep learning (xDL) in the nephropathology diagnosis process. This will probably entail looking into how xDL methods can deliver clear and understandable results, how they abide by important medical standards, and how they contribute to reliable and accurate diagnostic results for kidney diseases.

  • FDA

The FDA (U.S. Food and Drug Administration) is the regulatory body in charge of making sure that food, drugs, medical devices, and other products in the country are safe, effective, and of high quality. Your thesis topic, “Certification standards and Explainable Deep Learning (xDL) as basis in diagnosis process in nephropathology,” may have the following implications regarding the FDA:

  • Medical Device Regulation: It’s possible that your work may be governed by the FDA’s medical device regulations if it involves the creation or use of medical devices, software, or algorithms for using xDL to diagnose kidney diseases. The product may require FDA clearance or approval depending on its intended use and level of risk.
  • Regulatory Compliance: To guarantee patient safety and product efficacy, the FDA establishes standards and regulations for medical products. FDA regulations should be taken into account if your xDL-based diagnostic method is meant to be used for medical applications to make sure that it complies with the necessary standards.


Mask Region-Based Convolutional Neural Network (Mask R-CNN)

Mask R-CNN, or Mask RCNN, is a Convolutional Neural Network (CNN) and state-of-the-art in terms of image segmentation and instance segmentation. Mask R-CNN was developed on top of Faster R-CNN, a Region-Based Convolutional Neural Network. The first step to understanding how Mask R-CNN work requires an understanding of the concept of Image Segmentation. The computer vision task Image Segmentation is the process of partitioning a digital image into multiple segments (sets of pixels, also known as image objects). This segmentation is used to locate objects and boundaries (lines, curves, etc.).

There are 2 main types of image segmentation that fall under Mask R-CNN:

  • Vision Transformer (ViT)

Vision Transformers (ViT) is an architecture that uses self-attention mechanisms to process images. The Vision Transformer Architecture consists of a series of transformer blocks. Each transformer block consists of two sub-layers: a multi-head self-attention layer and a feed-forward layer. The self-attention layer calculates attention weights for each pixel in the image based on its relationship with all other pixels, while the feed-forward layer applies a non-linear transformation to the output of the self-attention layer. The multi-head attention extends this mechanism by allowing the model to attend to different parts of the input sequence simultaneously

Heat Map

A heat map is a two-dimensional color gradient used to depict data visually. It employs a color spectrum from warm to cold, with warmer hues denoting greater values and cooler hues denoting lower values. Heat maps come in many different shapes, but they always use color to communicate information. You can spot the areas on a heat map that have the most effects on raising a prediction’s likelihood by looking at it closely. Warmer regions, such as those shown in red, are typically more likely to have an impact on the forecast outcome, whilst cooler regions have a reduced influence


A number of metrics are relevant when assessing medical image understanding algorithms. The confusion matrix, also known as the error matrix, is an important metric. This matrix serves as a visual representation for evaluating the effectiveness of algorithms and calculating various evaluation metrics. Its goal goes beyond simple visualization; it explores the subtle mistakes the classifier made As a more complex and accurate metric by which to evaluate the performance of deep learning models, confusion matrices take the lead. Notably, the confusion matrix allows for four different possible combinations of anticipated and actual values when the model’s output spans multiple classes, providing a thorough picture of the model’s predictive abilities

  • When the projected value and the actual value are both positive, this is known as a True Positive (TP).
  • When both the prediction and the actual number are negative, this is referred to as a true negative (TN).
  • False positives (FP) occur when the result is erroneous even though the expectation was right.

False negatives (FN) are when the fact is positive but the prediction is negative.

Creation of ground-truth data

Customers  of the HS Analysis company worked together to arrange the specimens used for this study. The slides were scanned into NDPI files and then sent to HSA after being converted to digital form. The creation of deep learning (DL) models required these NDPI files.

The need for Ground Truth Data (GTD) is necessary to start DL model training. By creating a Base Region of Interest (ROI) and annotating the existing cells inside of it, this was accomplished. The HSA KIT proprietary software was used to meticulously create 8 files and more than 2,000 GTD instances for the study’s GTD dataset. There are several steps in the development of GTD. The initial loading of the Carl Zeiss Image Data File into HSA KIT (CZI), followed by the annotation of the kidney cell structures are annotated, specifically focusing on the “Tubules” structure.

ClassesBase ROITubules
All data142178
Used data81397


As with any other AI algorithm, a programming framework is required to create deep learning algorithms. These are usually extensions of existing frameworks, or specialized frameworks developed to create deep learning algorithms. Each framework comes with its own drawbacks and advantages. Let’s delve deeper into some of the most popular and powerful deep learning frameworks

Automated classification

The HSA software uses artificial intelligence (AI) to automatically classify images, allowing the detection of cells and/or classes through the use of Ground Truth Data (GTD), which forms the basis for model training. The effectiveness and efficiency of image analysis are improved by this method. Over 2,000 GTD instances and five files were used in this study, which involved two main implementation phases.

The initial model was trained using 5 files and the Mask R CNN which is based on the Mask-R-CNN architecture and focuses on tubules, in the first stage. The same process was then used, but with a slight modification in the second main stage: the Vision Transformer was used. The file with the GTD highlighted in red in Figure 20 denotes its detection as “Tubules.”

   Selection of the data set

After creation of GTD, the settings in Table 2 were used for 2 different architecture to train a model

Mask R CNN AND  Vision Transformer (VIT)

Model TypeEpochsLearning RateBatch SizeTile Size
Instance Segmentation1000.00012512   xDL technique and results

The HSA KIT’s heatmap tool was used to find picture regions specific to a particular class within this framework. Predictions are made after loading and preprocessing an image and running it through a model that has already been trained. The heatmap tool is used to integrate eXplainable DL (xDL) using this methodology.

The heatmap is then superimposed onto the original image to show the areas that affected the target class prediction. The tool performs its calculations, applies global average pooling, and generates the heatmap after the target class is specified. It is not necessary to have a large amount of Ground Truth Data (GTD) to view the visualization of AI detection methods in the heatmap tool, but more GTD enhances the heatmap visualization.Warm colors like yellow and red in Figure 21 denote areas within the detection area with a higher concentration or intensity of detection activity, as determined by the models. While Panel (B) shows the heatmap visualization, Panel (A) shows the absence of a heatmap.

1Interpretation of trained model results

In particular, this section will compare loss and accuracy evaluations for instances segmentation results obtained using both trained models (Mask R-CNN & Vision Transformer), specifically for Tubulus and the 1 classification. The actual outcomes of the model training are displayed in Table 3.

Mask R CNNTubulus0.57211195.713484
Vision TransformTubulus0.55157395.939921