Artificial intelligence in neuro-oncology: advances and challenges in brain tumor diagnosis, prognosis, and precision treatment

Abstract

This review delves into the most recent advancements in applying artificial intelligence (AI) within neuro-oncology, specifically emphasizing work on gliomas, a class of brain tumors that represent a significant global health issue. AI has brought transformative innovations to brain tumor management, utilizing imaging, histopathological, and genomic tools for efficient detection, categorization, outcome prediction, and treatment planning. Assessing its influence across all facets of malignant brain tumor management- diagnosis, prognosis, and therapy- AI models outperform human evaluations in terms of accuracy and specificity. Their ability to discern molecular aspects from imaging may reduce reliance on invasive diagnostics and may accelerate the time to molecular diagnoses. The review covers AI techniques, from classical machine learning to deep learning, highlighting current applications and challenges. Promising directions for future research include multimodal data integration, generative AI, large medical language models, precise tumor delineation and characterization, and addressing racial and gender disparities. Adaptive personalized treatment strategies are also emphasized for optimizing clinical outcomes. Ethical, legal, and social implications are discussed, advocating for transparency and fairness in AI integration for neuro-oncology and providing a holistic understanding of its transformative impact on patient care.

Fig. 1. AI-empowered multidisciplinary brain tumor management.

a AI augments the capabilities of neuro-oncologists/radiation-oncologists by enabling integrated diagnosis, offering deeper insights into the disease, facilitating precise prognosis by predicting outcomes, and assisting in patient stratification to tailor treatment plans to individual needs. b AI supports neuroradiologists by leveraging MRI images for automated detection and tumor segmentation, identifying molecular subtypes of tumors, providing quantitative measurements, and delivering diagnostic assistance to distinguish tumors from necrotic regions, all while ensuring automated quality checks. c AI aids neurosurgeons during surgery, contributing to surgical margin assessment and offering real-time diagnosis information and guidance, enhancing surgical precision and patient outcomes. d AI assists neuropathologists in the analysis of fresh/FFPE samples, providing automated measurement of features, aiding in tumor classification and grading, improving tumor detection, and delivering comprehensive analysis of cellular and tissue structures through histo-molecular classification. e Handling mutation data, single-cell information, methylation patterns, RNA sequencing, and more, AI empowers molecular pathologists by supporting biomarker identification, pathway identification, treatment response prediction, variant identification, and serving as a diagnosis assistant, streamlining the complex molecular analysis process (Created with BioRender.com).

Fig. 2. Multimodal integration for enhanced diagnosis, prognosis, and treatment response prediction in brain tumors.

Shown is the structural framework of a multimodal integration method designed to improve brain tumor management. The process involves the assimilation of data from five different sources, each contributing unique information. From MRI scans, radiomic data is generated. This data includes segmented MRI images achieved through AI-driven segmentation techniques, providing information about the tumor’s spatial characteristics. Blood samples yield ctDNA, allowing for the extraction of epigenomic, fragmentomic, and genomic alterations that inform the molecular landscape of the tumor. CFS samples provide cell-free DNA (cfDNA), offering genomic alteration information and contributing to a comprehensive understanding of the tumor’s genetic profile. Formalin-fixed paraffin-embedded (FFPE) tissue samples provide transcriptomic and molecular pathology data, offering information about gene expression and cellular structure. Clinical information such as age, race, gender, and electronic medical records (EMR) data supplement the molecular and imaging data, enriching the patient’s profile. For each of these modalities, feature extraction is performed, generating a set of informative characteristics. Subsequently, predictive models are applied to each dataset to estimate key outcomes related to diagnosis, prognosis, and treatment response. In the late multimodal integration, the predictions from these distinct models are fused to improve performance and precision. By synthesizing information from diverse sources and modalities, the integrated approach enhances the reliability and accuracy of neuro-oncological diagnosis, prognosis, and treatment response prediction (Created with BioRender.com).