Blogging is always one of the temptations, which helps in emulating the profession as a whole. It needs to be cognizant of what readers hope to derive from the blogs.
So What is Pathology?
It’s about diagnosing a disease, by analysing cells, tissues and body fluids or any sample of the body which helps in retrieving the cells. Routinely Haematoxylin and Eosin staining is done to appreciate the morphology of the cells. Special Stains like Silver Methamine, Jones, Perls, etc are used to stain specific structure. Further ancilliary techniques like Immunohistochemistry are applied to determine the right treatment option which helps in improving the patient survival.
What lies ahead of this Manual Pathology ??
The future of pathology in this era lies in Digital Pathology/Computational Pathology. Replacing the microscope with a Computer/Robot i.e.,machine learning of tissue and its analysis.
How computer is modelling an independent Human Pathologist’s mind ??
The answer lies in Artificial intelligence (AI) aka augmented intelligence is computer simulation of intelligent human behaviour.It helps the pathologist to analyse the digital slide image using image analysis system and Machine Learning/ Deep Learning algorithm, which can be supervised or unsupervised. It has improved the efficacy upto to 95- 99 %.
In a simple language, it aids the pathologist to diagnose, score or subclassify the type of cells in a particular patient type by using computer algorithms and interpreting them in the right way.
How AI implements in Pathology ?- A Brief Insight
It’s a 3 step process-
All the Histopathology/Cytopathology/Immunohistochemistry slides at local laboratories are scanned by using Whole Slide Image (WSI) analysis system.The data is sent to central cloud laboratory, which stores large data along with Electronic Health Record (EHR) of the patient. This process is called Digitalization or Digital Pathology. The interpretation of this data by sophisticated machines based upon various algorithms and machine learning approach is known as Pathology Artificial intelligence.
This involves segmentation, detection and classification as well as quantification and grading which helps the pathologist to form the specific Diagnosis (Dx), Prognosis
(Px) and Companion Diagnostics (CDx), which is the key to the health care system as shown in Fig1.
Assets of Artificial Intelligence above Manual Histopathology –
Manual histology provides single diagnosis, single scoring and single purpose.
While artificial intelligence provides the pathologists general purpose rich information data for tissue analysis, on which any number of single purpose interpretation or scoring schemes can be implemented.
Accuracy of automated morphological analyses has improved due to digital technology. AI is mainly used for cancer diagnosis, grading and pathological detection as of now, but now it has found its application in other subspeciality of pathology too.
Limitation of AI–
The key problem is the variation between the different patient types. No two patient look identical in disease state. And it is a tedious task to distinguish them based upon machine learning system.
Patient 1- Tumor cells are big in size, more compactly placed while the stromal cells are small and loosely placed and vice-versa in Patient 2.
Hence a lot of learning data is required so that the machine learning system would be able to learn the characteristic of cells of patient properly.
How to resolve and make it compatible for pathologist?
Whenever a new patient is encountered, a new patient type specific classifier is created. Then the pathologist trains the new classifier “on the fly” and all the cells get classified in that whole slide.
Over a period of time, multiple classifiers are fed into the system. Now when a patient with different cell characteristic is encountered, the best classifier is selected from all “patient type” specific classifiers, which then helps the pathologist with the best classification performance.
How much helpful is the Artificial Intelligent is to the pathologist??
Usually, pathologists agree to what they see in the tissue, based on their expertise, but have differences in how to get the overall diagnosis and score. For example- in a special IHC stain, a pathologist is asked to determine the percentage of stained cells of a particular type of tumor in a chosen compartment. This is a very challenging task which lead to high inter and intra observer variation.
So here comes the adoption of DIGITAL PATHOLOGY, which aids in improving accuracy and precision, reducing the pathologist’s burden and thereby enhances the efficacy.
LET’S PROVIDE THE PATHOLOGIST WITH THE RIGHT TOOL, REPLACE THE MICROSCOPE AND NOT THE PATHOLOGIST
Pathology Artificial Intelligence uses Artificial Neural network in machine learning for identifying the cell type, cell classification and complex scoring wherever needed, which helps not only in knowing the appropriate diagnosis, but is also very helpful in determining the patients prognosis and further treatment options.
Machine learning is based upon artificial neural network (ANN) having 3 functional layers. [Fig2]
Though microscopy is always the gold standard, but the limitation lies in the inter and intra observer variability and leading to error amongst the pathologist.
Lets take a quick example to understand the artificial intelligence application in Pathology.
Gleason system, one of the important prognostic factor in prostate cancer.And here comes the interobserver variability,so in order to get a consistent diagnosis, it is natural to introduce algorithmic intelligence in the pathology domain for the definitive morphological analysis.
Pitfalls of digital pathology –
Digital pathology do pose challenging problems like tissue handling,slide preparation, stain variation,image acquisition,different usage of algorithm,reporting and interpretations.In real life, AI cannot replace manual histology, though it yields high performance results.
Digital Pathology and Artificial Intelligence has gained its incorporation into genomics, proteomics and informatics(data-rich pathomics) leading to rapid development and prosperity of an AI-assisted pathology.
In this on-going pandemic, telemedicine and computer aided medicine are rapidly entering the markets in many countries. Despite the challenges, computational pathology contributes valuable insights to the diagnosis, prognosis, and treatment of disease ultimately.
…now is the time to create smarter healthcare systems in which the best treatment decisions are computationally learned from electronic health record data by deep-learning methodologies.” –Beau Norgeot, Benjamin S. Glicksberg and Atul J. Butte. Nature Medicine 2019; 25(January):14–18
Dr.Prachi
Senior Resident
Department of Pathology and Blood Bank
Dharamshila Narayana superspeciality hospital
New Delhi
Recent Comments