Artificial intelligence (AI) is rapidly transforming pathology and oncology, offering new opportunities to improve cancer diagnosis, prognosis, and treatment. This article reviews current applications of AI in tumor detection, molecular biomarker identification, and prognostic evaluation, as well as barriers and future prospects for its clinical adoption. AI, especially through deep neural networks, improves the detection and classification of tumors in histopathological images. Examples include the detection of lymph node metastases in breast cancer and the accurate classification of tumor and normal tissues in multiple types of cancer. AI algorithms can identify traditional and novel biomarkers from digital images, reducing inter-observer variability and optimizing analysis time. This includes the prediction of mutations, molecular subtypes, and tumor microenvironment characteristics. AI models analyze clinical, histological, and molecular characteristics to predict treatment response and survival. Some systems have identified histological patterns not previously associated with prognosis. Technical, regulatory, and cultural challenges remain, including data availability, model construction and understanding, process standardization, and rigorous clinical validation.
The integration of multimodal models, the use of advanced architectures such as transformers and foundation models, and international collaboration to expand information are key to advancing toward more equitable and effective precision medicine. Introduction Oncology is undergoing a profound transformation driven by the emergence of artificial intelligence, especially the development and application of deep learning (DL) algorithms in the interpretation of histopathological images. Since the 2010s, advances in computing power, the massive availability of high-resolution digital data, and improvements in neural architectures have led to a qualitative leap in the ability to automate, optimize, and standardize diagnostic processes in oncology. This article systematically reviews the scientific literature on AI applications in oncological pathology, with an emphasis on three areas: 1. Tumor detection and classification based on whole-slide images (WSI). 2. Identification and development of molecular biomarkers for diagnosis, prognosis, and therapeutic selection. 3. Prediction of prognosis and response to treatment in cancer patients. It also addresses technical, regulatory, ethical, and cultural barriers that hinder clinical implementation and examines the future prospects of this technology in the context of precision medicine. Results Applications in tumor detection and classification Sixty-eight studies (2013–2022) were identified that applied AI algorithms, mostly convolutional neural networks (CNNs), for automated diagnostic tasks. These applications include: n Localization of neoplastic regions. n Classification of lesions as benign or malignant. n Determination of histological subtypes and tumor staging. A paradigmatic example is the CAMELYON16 competition, which demonstrated that DL algorithms could match and, in certain scenarios, exceed the performance of human pathologists in detecting lymph node metastases in breast cancer. Other studies have extended this approach to common tumors such as breast, prostate, lung, and gastrointestinal, achieving areas under the curve (AUC) greater than 0.90. The PC-CHiP model, based on CNN Inception-V4, stood out for its ability to correlate histological patterns with genomic alterations, such as genomic duplications and specific mutations, as well as for its potential to predict survival in multiple tumor types.
COMMENTARY BY DR. RODRIGO SÁNCHEZ BAYONA, SCIENTIFIC SECRETARY OF SEOM
Authors A. Marra, S. Morganti, F. Pareja, G. Campanella, F. Bibeau, T. Fuchs, M. Loda, A. Parwani, A. Scarpa, J. S. Reis-Filho, G. Curigliano, C. Marchiò & J. N. Kather
To download the complete research, click below:
Page 61
https://seom.org/images/REVISTA_159_PLS_SIN_ANUNCIOS.pdf
Join the doctors who get the most relevant medical news, curated by specialty, every week — no spam, unsubscribe anytime.
