Challenges and perspectives in use of artificial intelligence to support treatment recommendations in clinical oncology

Abstract

Artificial intelligence (AI) promises to be the next revolutionary step in modern society. Yet, its role in all fields of industry and science need to be determined. One very promising field is represented by AI-based decision-making tools in clinical oncology leading to more comprehensive, personalized therapy approaches. In this review, the authors provide an overview on all relevant technical applications of AI in oncology, which are required to understand the future challenges and realistic perspectives for decision-making tools. In recent years, various applications of AI in medicine have been developed focusing on the analysis of radiological and pathological images. AI applications encompass large amounts of complex data supporting clinical decision-making and reducing errors by objectively quantifying all aspects of the data collected. In clinical oncology, almost all patients receive a treatment recommendation in a multidisciplinary cancer conference at the beginning and during their treatment periods. These highly complex decisions are based on a large amount of information (of the patients and of the various treatment options), which need to be analyzed and correctly classified in a short time. In this review, the authors describe the technical and medical requirements of AI to address these scientific challenges in a multidisciplinary manner. Major challenges in the use of AI in oncology and decision-making tools are data security, data representation, and explainability of AI-based outcome predictions, in particular for decision-making processes in multidisciplinary cancer conferences. Finally, limitations and potential solutions are described and compared for current and future research attempts.

Keywords: artificial intelligence; clinical oncology; genitourinary cancer; multidisciplinary cancer conferences; treatment recommendation.

FIGURE 1

Extension of the medical process by an AI‐supported process, which enables medical staff to support diagnosis and treatment recommendation. This shows an example of a process in which the medical process of generating a treatment recommendation is supported by an AI process. The basic idea is to perform validation after the process, which minimizes the risk of error in the recommendation. First, investigations are performed to collect patient data. This can be any type of relevant data. Next, evaluation is performed by medical professionals and a recommendation is made. The patient data would be evaluated using an AI while maintaining privacy. The AI makes a recommendation, which is presented to the healthcare professional‐ Ideally, an explanation should be generated by the AI that allows the recommendation proposal to be understood. After the evaluation by the specialist, a final recommendation can be made by the medical professionals, which might include suggestions or be supported by the AI‐generated recommendation.

FIGURE 2

Artificial intelligence represents of the major current research fields in clinical medicine, in particular due to its opportunities in supporting diagnoses and treatment recommendations. This technical support might help to achieve a healthier global population and might have a positive impact on various application fields, such as resource optimization, efficiency, remote monitoring, and others.

FIGURE 3

Machine learning can be deployed in various application fields. However, it requires feature engineering. Feature engineering is a complex approach that requires understanding of the data, which can be problematic in some cases. Deep learning can work with almost all kind of data and in addition do not require a costly feature engineering, making it possible to quickly explore the data without prior knowledge.

FIGURE 4

Illustration of different working steps required to develop a trustful system for medical use cases. First (1), a reliable data extraction is required to gather information from different sources and standardize it in a way that is readable for the machine learning. Second (2), an inference system needs to be developed and optimized for the use case. To achieve the best result, multiple approaches need to be tested, optimized and possible combinations need to be evaluated. Finally (3), the accuracy and explainability need to be evaluated including further analysis by experts and a vulnerability analysis to ensure a reliable and trustful system.

FIGURE 5

Illustration of different concepts used in attribution methods to produce a “heat map.” Gradient‐based approaches depend on the loss and require a backward pass of the data to mathematically compute the importance value of each point. Permutation‐based approaches only require forward passes and change the input to understand the impact of input changes. Replacement‐based approaches are related to the previous mentioned approaches but are different in the way that they do not perform a small change of the point but rather completely remove the point or set it to a fixed value.