Dr. Mona Jamwal
Dr. Poonam Sharma
Dr. Reetika Menia
John McCarthy introduced the term “artificial intelligence” in 1956, and Arthur Samuel coined “machine learning” in 1959, defining it as the ability to learn autonomously. Artificial intelligence (AI) is becoming increasingly important in the medical field. The term AI refers to intelligence provided by non-human artificial entities, such as computers and robots in pathology.
The role of AI in pathology: AI and machine learning technologies are improving the efficiency and accuracy of diagnosing diseases such as cancer. Digital pathology, which uses advanced image processing and AI algorithms, is becoming more common and provides faster and more accurate diagnoses. AI's ability to effectively analyze complex data is helping improve precision medicine for cancer patients. Automated whole-slide imaging scanners produce high-quality images of tissue samples, which, when combined with digital pathology tools, streamline the diagnostic process across various areas of pathology. Notably, in 2018, the FDA approved the first medical device to utilize AI to detect diabetic retinopathy in adults. Then, in 2021, the FDA cleared Paige Prostate, the first AI-based software to detect prostate cancer. Machine learning (ML) techniques, a subset of artificial intelligence (AI), are increasingly employed in pathology to automate image analysis tasks and improve diagnostic accuracy. These methods cover a wide range of functionality, from basic object recognition to complex pattern analysis for disease diagnosis, prognosis, and predicting treatment outcomes. By extracting image patches and leveraging deep learning algorithms, AI enables detailed spatial examination of histological features for faster and more reliable diagnosis. Additionally, AI streamlines tasks, allowing pathologists to focus on complex cases and meet increasing workload demands. Identify areas of interest, prioritize cases, and introduce new staging paradigms to improve workflow efficiency. Deep learning, the foundation of AI and ML, offers great potential to improve diagnostic capabilities in pathology by using neural networks to process data and make predictions.
AI in pathology has two uses. Clinically, it streamlines tasks such as cell counting and cancer detection, increasing accuracy and saving pathologists time. Additionally, it also serves as a quality control tool, increasing the reliability of diagnostic reports. In research, AI mines complex medical data to uncover new disease patterns that are important for customized treatments and prognosis. For example, it accurately predicts the prognosis of kidney cancer and outperforms traditional methods. AI is also great at predicting treatment response for colorectal cancer patients and identifying biomarkers in digital slide images, playing a pivotal role in the evolution of pathology and advances in treatment.
Application of AI in Diagnosis: Recent advances in AI have revolutionized the diagnosis and classification of cancer, especially in pathology. AI tools integrated into diagnostic workflows tackle a variety of tasks such as object recognition and segmentation to help pathologists identify complex information. For example, AI increases sensitivity by efficiently detecting isolated tumor cells in lymph nodes suspected of metastasis. Scoring criteria have also been standardized in tumors such as prostate and breast cancer, where the morphology indicates a biological process. Content-based image retrieval (CBIR) allows pathologists to search vast databases for images similar to a particular case, which is critical for accurately and quickly diagnosing rare and complex conditions. This technique highlights similarities in histopathological features and facilitates diagnosis of difficult cases beyond visual similarities.
Predictive and prognostic applications of AI: AI has shown promise in predicting prognosis and treatment response based on histological features, efficiently providing important insights. By correlating images with tumor characteristics, microenvironment, and genetic profiles, AI can provide concise predictions about survival and treatment outcomes. Integrating diverse morphological features into a unified prognostic indicator is a challenge for humans, but AI-based image analysis provides a new classification system. This system represents clinical outcome, likelihood of recurrence, and treatment response by correlating important histological features such as tumor morphology and stromal structure. Deep learning methods within computational pathology (CPATH) discover distinctive biomarkers, such as tumor-infiltrating lymphocytes (TILs), whose spatial distribution holds diagnostic and prognostic value. TILs, which are important for the activation of anti-cancer immunity, have the potential to serve as reliable biomarkers if they can be objectively quantified throughout the tumor microenvironment (TME).
AI as a predictor of molecular and genomic profiles: AI tools play a key role in extracting insights about tumor genetics and genomic profiles from morphology, helping to understand cancer biology. Molecular-based tests, like mRNA-based tumor type tests, provide prognostic information by combining different parameters. Although linking morphological patterns to tumor genetics seems straightforward, integrating vast amounts of genomic data, such as from next-generation sequencing (NGS), presents challenges. Although combining images and molecular features can provide a comprehensive view of tumors, developing, training, and validating models to handle such complex multidimensional data remains a major challenge.
The role of AI in research, training, and education: AI tools provide interactive features and annotations to create dynamic learning environments and are valuable resources to enhance pathologist training. This integration will enhance morphological knowledge with advanced technology and enable personalized precision medical practice. AI tools are already being used in educational settings such as conferences and virtual workshops, and alongside clinical and research registries and advanced laboratory information systems, provide comprehensive support in pathology practice. . By integrating AI into daily workflows, trainees receive supplementary information for differential diagnosis and improving diagnostic skills, training residents and collaboration between institutions for efficient consultation of difficult cases. strengthen.
Future Directions and Opportunities: The development of AI tools for cancer detection by various companies has proliferated in recent years, and the FDA's approval of the Philips whole slide scanner in 2017 is an important step in the digitalization of clinical workflows. It is an important milestone. The COVID-19 pandemic has prompted some educational institutions to adopt digital workflows, despite the challenges. Despite the obstacles, there are significant changes in digital pathology, with the introduction of open-top light-sheet microscopes that provide non-destructive, slide-less 3D tissue images, enabling AI applications with better spatial and structural information. may be strengthened. While current AI can identify tumor scores and grades, future applications are likely to continue to focus on narrow AI tasks.
Over the next decade, AI tools are expected to become commonplace in pathology laboratories and are likely to be utilized for a wide range of tasks. These tasks include identifying micrometastases and lymph node metastases, quantifying various stains, counting mitosis and lymphocytes, and automating slide and image quality control processes. Additionally, AI is expected to play a key role in prioritizing cases, standardizing pathology reports, and assisting in laboratory workflow management. In recent years, new AI approaches in pathology have shown great potential to improve diagnostic workflows, reduce errors, and enhance prognostic prediction. However, challenges such as interpretability, validation, regulation, and cost have slowed their integration into clinical practice. Integrating AI with human pathologists and existing systems could be beneficial to meet the growing demand for personalized cancer treatments. Multimodal approaches that combine proteomics, genomics, and AI-based biomarker quantification may be important for precise tumor therapy.
(Author Dr. Mona Jamwal is a senior resident in the pathology department at AIIMS Jammu, Dr. Poonam Sharma is an associate professor and Dr. Reetika Menia is an assistant professor).
