Could pathologists soon be replaced by robots?

Abstract

Lou Gerstner, a prominent American businessman, once said, “computers are magnificent tools for the realization of our dreams, but no machine can replace the human spark of spirit, compassion, love, and understanding”. Although not stated in the context of this contention, this quotation appositely summarizes a consensus toward robots and their potential in the medical profession, namely, they lack the capacity to replace humans outright. We aim to advance this notion, ultimately concluding that robots will not supplant pathologists, but should instead be deployed in conjunction. Pathology involves understanding the cause of disease, the processes inherent in testing for disease, and its reporting. This may involve investigating changes in tissues, blood, and body fluids to devise a definitive diagnosis which can be used to guide treatment. Given the data-driven nature of some aspects of this work, it makes sense that robots are considered alternatives to pathologists. But before we delve into applications of these machines, we must first establish whether there is a need for robots in the pathology industry. Superficially, there appears to be little need to change the current state of pathology practice. A customer satisfaction survey reported the median overall satisfaction score for the performance of pathologists was 4.4 of 5, among the best of all medical specialities.[1] However, upon closer inspection, it becomes apparent that there is room for improvement. One example is the issue of significant interobserver variation when reporting tissue slides, which has been estimated at 18.3% across different fields of pathology.[2] This phenomenon is prevalent in day-to-day practice and misdiagnosis can lead to delay in treatment, ineffective interventions, and reduced overall survival. A related limitation is that the work of a pathologist is under the influence of their emotional and/or physical state, which can introduce inconsistencies. These problems add to the fact that demand for pathology services has grown significantly in recent years, but staffing has not experienced a commensurate increase. Given these limitations, improvement in the industry is necessary. Robots have increasingly been perceived as a convenient solution and have already been implemented in a clinical setting. The term “robot” in this context refers to any automated algorithm capable of assisting pathologists in their role. A common feature of such algorithms is their utilization for artificial intelligence (AI), that is, using mathematically derived computational methods to make predictions and emulate human decision-making. There are three main ways in which AI is currently being used in pathology—replacement, alternative triage, or augmentation—and their workflows are summarized in Figure 1. Robots are already used in pathology. For instance, Yu et al. (2016) designed fully automated machine-learning methods for survival prediction of lung adenocarcinoma and squamous cell carcinoma patients from histopathology images. The validated survival prediction proved significantly more accurate than the current ‘human’ standard of practice.[3]Figure 1: AI can be incorporated into the pathology workflow in three main ways: (a) replacement; (b) triage; and (c) augmentationThe application of robots and computational techniques in lung cancer diagnosis is one example of the many uses of robots in pathology. However, what we should evaluate is the extent to which robots are, indeed, replacing pathologists as opposed to simply complementing their capabilities. In doing so, we find that there are some aspects of a pathologist’s role that are wide-ranging and irreplaceable [Figure 2].[5] For one, robots, at least in the foreseeable future, will not be able to communicate and engage with patients and colleagues to the same extent to which pathologists do. Furthermore, pathologists have traditionally played important roles in research. We need to only look as far as Nobel Prize Laureates Dr. Robin Warren and Dr. Barry Marshall, whose work underpins treatments for Helicobacter pylori-associated peptic ulcers, to see that the work of pathologists has been clinically valuable. Pathologists remain essential, with 89% of UK pathologists contributing to research at some point during their career—an asset we would forego by replacing pathologists with robots.[4]Figure 2: Breakdown of typical activities engaged by pathologists in proportion to the average time spent on each activity—Modified from Trotter et al. (2019)[ 5 ]Evidently, it is not feasible for robots to entirely replace pathologists. A more likely scenario is for humans and machines to work in conjunction. Such implementation was demonstrated by Wang et al. (2020) who designed a system to facilitate collaboration between expert pathologists and an automated detection system used to identify metastatic breast cancer on whole slide images of lymph node biopsies. This method significantly increased diagnostic accuracy, representing an 85% reduction in the human error rate.[6] Paige.AI prostate is an AI-powered digital pathology platform designed to assist pathologists in detecting and grading prostate cancer on digitized biopsy slides.[7] The platform has received Food and Drug Administration clearance in the United States and is one of four healthcare AI-pathology algorithms currently approved.[8] From detecting acute leukemia in peripheral blood[9] to tissue processing of hepatocellular-cholangiocarcinoma using cloud-based AI technology,[10] machines will likely help us reach our fullest potential. So, what are the challenges ahead? There is a myriad, but among the most pressing is the availability and cost of either generating or obtaining standardized data in large quantities that may be needed to train AI algorithms that form robots. This is likely to pose financial and logistical problems particularly in low-income settings where expertise in building AI tools alone may pose a barrier. In addition, there exist questions surrounding a medico-legal responsibility for diagnostic errors made by a human/AI combination. Does the responsibility fall upon the platform, the pathologist, or both? In general, doctors operate under the Bolam guidelines, which state a doctor is not negligent if he or she has acted in accordance with a reasonable body of medical professionals. It remains unclear how these regulations would apply to errors made by robots. For these reasons, until conciliation is obtained, ambiguity surrounding how robots can and should be implemented in the clinical setting will persist. The medical industry finds itself at an exciting crossroad in deciding whether to pursue mechanization or remain grounded in traditional practice. At the time of writing, alongside “traditional” AI tools to help solve inference and prediction problems in healthcare, several exciting generative AI tools are now being released. These also have wide-reaching implications in pathology, from potentially helping doctors synthesize diagnosis notes to generating synthetic images for training purposes. There is little doubt that robots are becoming increasingly capable and adaptable, particularly within pathology. However, it would be naïve to consider robots as a replacement for pathologists. Instead, these machines should be viewed as a counterpart to augment a pathologist’s role. We ought to now turn our attention to how best to achieve this symbiosis in everyday practice. Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest.

Ähnliche Arbeiten