Role of AI and Machine Learning in Hematology

Introduction:

Despite increasing recognition of the importance of genetic status, hematologic diagnosis is still primarily based on assessing phenotypic traits. Until recently, all research on prognostic markers and treatments for hematologic malignancies was based on microscopically defined disease entities according to World Health Organization (WHO) classification guidelines. Nonetheless, diagnostic ambiguity occurs regularly, and the quality of results is highly dependent on the experience and skill of the operator. To reduce reliance on expertise and potentially increase the consistency of data interpretation, various downstream artificial intelligence (AI). It is preferable to implement an automated process to support the development project.

What Is Artificial Intelligence?

In its simplest form, artificial intelligence is the field that combines computer science and robust data sets to enable problem-solving. This also includes the machine learning and deep learning sub-areas often mentioned in connection with artificial intelligence. These areas comprise AI algorithms that are aimed at creating expert systems that make classifications based on input data. Over the years, artificial intelligence has gone through many hype cycles, and so have some. Language generative models can also learn grammar for software code, molecules, natural images, and many other data types. The applications of this technology are increasing day by day, and it is just getting started.

Weak AI (that is also known as narrow AI) is a type of AI that is trained and works intensively to perform a specific task. Weak AI underpins most of the AI ​​that surrounds us today. 'Narrow' might be a more accurate term, as this type of AI is by no means weak.

Strong AI comprises artificial superintelligence (ASI) and artificial general intelligence (AGI). Artificial General Intelligence (AGI) or General AI is a theoretical type of AI in which machines exhibit human-like intelligence. One will have a confident mindset and the ability to solve problems, learn, and plan for the future. Artificial Super Intelligence (ASI), also called superintelligence, will surpass the intelligence and capabilities of the human brain.

What Is the Role of AI and Machine Learning in Hematology?

Medical imaging has undoubtedly benefited most from the introduction of AI-powered scoring systems, which have been successfully applied to cancer detection for many types of cancer. In hematology, AI-based methods automatically recognize and classify single-cell images or superimposed patterns from digital microscopy and support the evaluation of flow cytometry data used to assist the cytogeneticist. Karyotyping is used to develop individualized models that integrate multiple data sources to estimate the presence or absence of response to specific treatments and accurately predict the prognosis of various leukemias.

What Are the Uses of Machine Learning for Hematologists?

To solve a problem computationally, an input (microscopic image, mutation profile) is transformed into a desired output (cell type classification, prognostic score) according to a set of instructions (algorithms). Needless to say, cell type classification requires different manipulations than automated interpretation of genetic variation. However, for most tasks in hematology, it is very tedious to program all the rules and exceptions rigorously. For example, defining rules for the automatic classification of blood cells quickly becomes tedious. In contrast, for other tasks, such as the interpretation of genetic mutations, knowledge is lacking, and a comprehensive set of solutions to solve the problem is required. This is where machine learning (ML), a subset of AI, comes into play. Generate predictive models that learn directly from observed data without requiring explicit instructions.

What Are the Various Methods of Machine Learning?

In the field of medicine, the terms 'machine learning' and 'artificial intelligence' are often used interchangeably. However, ML refers to the automatic recognition of patterns and relationships in data, while AI aims to simulate human behavior and intelligence. Broadly speaking, there are two categories of AI:

• Strong AI - A type of system that can find viable solutions to various problems without the need for human intervention.

• Weak AI - It is implemented to perform a single task based on a specific data set within a predefined range and is relevant to today's medical ML-based algorithms.

Various types and techniques of ML are used in hematology. Hidden model reasoning is central to ML, and three main methods are:

• Supervised Learning - Supervised learning is the most commonly used technique for building clinical models that help hematologists make diagnostic, prognostic, or therapeutic decisions. As you train the model, it iteratively improves its performance by matching predictions with labeled data. A model's accuracy is evaluated by evaluating its performance on the holdout dataset.

• Unsupervised Learning - Unsupervised learning can be used for exploratory analysis of unlabeled data to infer underlying factors and their interactions. Unlike supervised learning, there are no predefined outcomes, and assessing the value of groups identified by associating them with clinical and/or genetic factors requires a manual assessment of outcomes.

• Reinforcement Learning - The learner or agent aims to maximize the expected utility, similar to the situation where a clinician needs to adapt their behavior (treatment) to the patient's condition (genetic profile). It also performs various operations to interact with the environment adaptation. Therefore, RL can be used to optimize treatment for patients with different characteristics, and the effective combination of RL and deep learning (DL) can further improve model performance.

What Is the Future Direction of AI and Machine Learning in Hematology?

AI-based technology has evolved rapidly over the past five years, giving rise to a number of narrow AI applications that can be used at every stage of patient management in hematology, from the analysis of peripheral blood differentials to genetic profiling. Combining AI-based models with human doctors can advance hematology diagnostics in ways beyond what they can do alone. The implementation of ML techniques will, on the one hand, assist clinicians in analyzing and interpreting data and increase the objectivity and accuracy of work-up, but on the other hand, the accumulated and integrated knowledge will be an integral part of the experience. It helps doctors guide their practice in the decision-making process.

However, before understanding the benefits of AI, one must also consider the ethical challenges of implementing machine learning in the medical field. Some ethical issues are straightforward, such as concerns that algorithms may reflect human biases in decision-making and reinforce existing social biases. Other ethical issues include:

• Changing relationships between physicians and patients.

• Protecting patient privacy.

• Physician-patient trust in ML-assisted decision-making.

ML-based techniques can also be trained in unethical ways, potentially increasing developer profits or improving quality metrics without providing better patient care.

Conclusion:

AI-based methods will not replace clinicians in the near future, but they can support hematologists' decision-making by professionally completing tedious, repetitive, and time-consuming tasks, making it possible for each patient. It should be used to ensure the best possible diagnosis. Extensive training is therefore required to provide the next generation of scientists and physicians with the necessary background knowledge to properly evaluate new developments, manage their integration into the healthcare system, and successfully apply them.