Introduction:
Pathology is a branch of medical science that helps identify different human diseases and causes, specifically focusing on the structural, morphological, functional, and molecular changes produced in cells and tissues. To make the process easier and more universal, computational and systems pathology (CSP) will be helpful. It has a similar scope and the scientific goals of classical pathology disciplines.
The main characteristics of CSP are the utilization of computational approaches and mathematical modeling to know about diseases. In addition, it also provides many data obtained from different sources like digital images, molecular data, etc. The aim of the CSP is to improve patient care and outcomes by speeding up the diagnostic processes.
What Is Computational Pathology?
Computational pathology is a branch of pathology that aids in diagnosis by combining data from multiple sources, such as pathology, clinical, radiology, and molecular, using machine learning (ML) and mathematical models. This helps develop the best possible decision by enhancing diagnostic accuracy and lab efficiency. It helps to examine the slides virtually with the respective data, which includes patients' medical and insurance information, how was samples acquired, and annotations.
It is very efficient with its deep learning algorithms and convolutional neural networks to fetch information from visual imagery, making them a perfect constituent in computer vision and image recognition in healthcare. Artificial intelligence (AI) also plays a paramount role in collecting information that is overlooked or invisible to human eyes due to factors like negligence, weariness, etc. AI improves the proficiency of tissue-based diagnosis, offering more sensitive and accurate readings than human pathologists. Additionally, it also plays a dominant role in precision medicine, health care designed to advocate personalized treatment.
What Is the Role of Computational Pathology in the Modern World?
Computational pathology goals are similar to classical pathology disciplines. The unique aspect of computational pathology is using mathematical modeling and computational approaches and knowing about the disease process. This is carried out by collecting data from multiple sources, such as digital images, molecular data, etc. The aim of the long series of effects is to generate precise diagnostic processes and treatments to improve patient care and outcomes. The scope of computational pathology prioritizes comput0ational methods and understanding the disease pathogenesis, and offering treatment. It is also paying attention to developing the core competencies of pathology to expand the ability to provide actionable knowledge effectively. The main areas of science focused on computational pathology are clinical genomics and other omics, digital pathology and image analysis, and the integration of multi-modality data with clinical information.
What Are the Limitations of Implicating Computational Pathology?
The transition from normal pathology to computational pathology is still a faraway project. This can occur in the future by incorporating a pathologist's supervision and computerized automation. There are many limitations and challenges in establishing a computational pathology; a few are mentioned below.
1) Multi-Gigapixel Images - Implementing computational pathology depends solely on powerful hardware and effective software to handle large gigapixel WSIs (whole slide imaging). Because of the large size of the image, there comes the need for significant storage in the pathology department, and also transferring data is a backbreaking task. To overcome this, a patch-by-patch approach is followed, which allows breaking down the images into smaller patches or tiles on annotation input or other preprocessing steps. The patching is a time-consuming process and also needs humongous storage. Patching has the trouble of contextual information loss and computational constraints that makes compromises with the resolution of the images and patch size.
2) Standardization - The quality of the WSI is determined by the quality of slide preparation, which involves processes like tissue fixation, cutting the slices, and staining the histological slides. Also, the scanning process may be carried out with different types of scanners and settings. All these different steps of processing may cause variations in WSI and make automatic processing a difficult task. WSIs can identify and exclude minor errors while processing the slides, such as issue folds, air blobs, shadows, and blurring, where the quality of the images is compromised and reduce the performance of systems that analyze them. Also, in some cancer cases, the tumors are removed by cauterization, damaging the tissue part and losing needed diagnostic information.
There will be variations in the staining colors of WSI images inter-hospital due to different protocols being followed, like the stain used and acquisition system differences. Also, intra-hospital variability can happen due to differences in the thickness of tissue slides, lack of uniformity of the stains used, age of the slides before scanning, etc. These color variations put the analogous outcome in deep learning models in difficulty. Because it is impossible to provide a diverse training set with all possible normal variations, different approaches should be followed to deal with variations of colors of the WSIs, like color normalization and augmentation.
3) Labeling - Artificial intelligence algorithms are completely data-based and need different data to produce good AI models. AI models can be well-trained by supervised, semi-supervised, and unsupervised techniques. Supervised learning is simple and the most straightforward strategy, where data has truth labels to tune the model's parameters. To have a clinically relevant label, pathologists are required to label the WSIs. Also, both semi-supervised and unsupervised methods require a test set labeled according to the task and performance of prediction models.
4) Data Fusion - To evaluate the behavior for carcinoma, only the present clinical and histopathological features are not adequate. Combination data is necessary for personalized diagnosis and treatment. Different noninvasive techniques, such as endoscopy, ultrasound imaging, magnetic resonance imaging (MRI), and optical coherence tomography (OCT), are also widely performed for the early detection of the disease and evaluation of the progression of a tumor. Mostly all these procedures are usually done before the biopsy and, therefore, before the histopathological examination; its results provide pathologists with details of regions and features of interest. So, combining WSIs with other data may be relevant in the future of computational pathology, helping professionals in the detection, diagnosis, treatment, and prognosis. Data fusion can be implicated in early and late fusion modalities.
Conclusion:
Computational pathology is a revolution in classic pathology that can provide automatic diagnostics, involve pathologists with specific areas of interest, and predict prognostic. By implicating such new technologies, it is believed that the workload of pathologists and the turnaround time at pathology labs will be reduced. So this area is getting more limelight and becoming a significant area of research and having rapid growth. However, there is many cumbersome to overcome and bring computational pathology into action, which is concentrated by many research community from different angles to come up with solutions for each problem in all applications.
