Can Artificial Intelligence Replace Humans in Histopathology?

    Convolutional neural networks (CNNs) have been an important component in the development of artificial intelligence (AI) and machine learning. It has been used in interpreting visual and spatial data and is being trained to perform in tasks related to image recognition, language processing, and tasks in the medical field (ScienceNewsToday). Delving into the medical field, pathology is a specialty that involves histology, which means studying tissue samples under a microscope to determine if the tissue is diseased. As artificial intelligence such as CNNs continue to evolve, it begs the question, can they replace humans in histopathology?

    During Dr. Baker’s talk at Loyola, he highlighted his own research regarding how deep convolutional models (DCNNs) struggle with the configuration of images compared to humans. In his paper, Dr. Baker highlighted how DCNNs focused more on color and texture than humans do (Baker). Additionally, in his presentation he went into detail into how the AI struggled with recognizing an image based on manipulation of the images’ border. For example, if the AI was given the silhouette of a cat, but if the border was turned into rigid edges rather than the normal “smooth” border, it would have more trouble recognizing what the image was in comparison to an image that was “Frankensteined” (chopped up and put together in a different configuration).

    It is necessary to make note of this distinction in the AI’s recognition of the image because cell tissues do not always maintain a constant shape or look, meaning that Dr. Baker findings could pose an issue of accuracy when using AI in pathology. A review article by Prasad et al. delves further into the use of AI in the field of histopathology. In their findings, CNNs were found to have consistent results in detecting and classifying cancerous tissue and had levels of accuracy similar to trained pathologists in controlled settings (Prasad). However, the article highlights an issue in which AI systems struggle with variability in staining, slide preparation, and tissue morphology (Prasad).

    This connects to Dr. Baker’s findings, as the AI relies heavily on texture and color rather than truly understanding structural features, which limits the effectiveness of CNNs when analyzing histological samples that are inconsistent or irregular (Baker). Furthermore, Prasad et al. highlight that while AI can assist with lowering workloads and increasing efficiency, as mentioned, it still lacks the adaptability needed to analyze irregularities that can be common in complex histopathological cases (Prasad). This shows that while AI can become an assistive tool in histopathology, its limitations in image interpretation demonstrate that humans in pathology are still necessary. However, rather than replacing pathologists, AI is more useful as a supportive tool that enhances accuracy and efficiency.

 

References

Editors of ScienceNewsToday. (2026, April 7). Convolutional neural networks: the science behind modern artificial intelligence. Science News Today. https://www.sciencenewstoday.org/convolutional-neural-networks-the-science-behind-modern-artificial-intelligence#google_vignette

Baker N, Elder JH. Deep learning models fail to capture the configural nature of human shape perception. iScience. 2022 Aug 11;25(9):104913. doi: 10.1016/j.isci.2022.104913. PMID: 36060067; PMCID: PMC9429800.

Prasad P, Khair AMB, Saeed M, Shetty N. Artificial Intelligence in Histopathology. J Pharm Bioallied Sci. 2024 Dec;16(Suppl 5):S4226-S4229. doi: 10.4103/jpbs.jpbs_727_24. Epub 2025 Jan 30. PMID: 40061791; PMCID: PMC11888715.