Image by vecstock on Freepik

Background

OC is the most common malignant tumor in the female reproductive system. Despite treatment advances, 70% of late-stage OC patients relapse post-treatment, with very low survival rates. In 2020, approximately 314,000 new OC cases were diagnosed globally, with around 207,000 deaths.

Recently, AI-based multi-omics research has been extensively conducted, focusing on OC diagnosis, differentiation of benign and malignant tumors, pathological classification, drug efficacy, and prognosis prediction. Researchers have reviewed AI's clinical applications in OC. However, previous studies mainly discussed image-based content without detailing other omics techniques like pathology, genomics, and transcriptomics. Additionally, most studies only evaluated AI's value in single-omics applications. This study provides a comprehensive review of AI's applications in radiomics and other omics.

Radiomics in Ovarian Cancer Diagnosis

Radiomics is a non-invasive method that extracts high-throughput image features from medical images, related to tumor pathophysiology. In OC diagnosis, radiomics has proven to be a convenient and cost-effective tool, valuable for diagnosing gynecological diseases, tumor staging, genotype prediction, and prognosis assessment.

The main steps in radiomics include medical image acquisition, image segmentation, feature extraction, feature selection, and model construction. Common imaging methods include computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), and ultrasound. Image segmentation typically involves automatic, manual, and semi-automatic methods to determine regions of interest (ROI). Feature extraction covers tumor morphology, first-order, second-order, and higher-order features, reflecting tumor shape, size, vascular distribution, and its relationship with surrounding tissues.

In OC radiomics research, AI has begun to standardize and simplify the diagnostic process. AI can mimic human cognitive behavior, extracting information from images for feature selection and model construction. Machine learning (ML), the core of medical AI, includes supervised learning, unsupervised learning, and reinforcement learning. Deep learning (DL), a subset of ML, solves complex problems through multiple artificial neural networks, automatically identifying features in data, avoiding manual feature selection .

Radiomics, ML, and DL are not independent but often intricately intertwined. The radiomics modeling process typically relies on DL. Currently, AI is widely used in disease diagnosis, differentiation of benign and malignant tumors, and treatment effect prediction.

Predicting Different Pathological Subtypes

AI technology assists doctors in more accurately classifying OC subtypes by analyzing medical imaging data. For example, AI models can distinguish between Type I and Type II EOC through ultrasound image analysis, crucial for selecting appropriate treatment plans and predicting treatment outcomes. Studies show that ultrasound-based radiomics models have high accuracy in predicting different pathological subtypes of EOC, using LASSO regression to select key features and establish models with satisfactory predictive efficiency.

Predicting Gene Mutation Status

AI also shows great potential in predicting gene mutation status. Since OC patients' treatment response is closely related to gene mutation status, such as BRCA gene mutations increasing sensitivity to platinum-based drugs, accurately predicting gene mutation status is crucial for personalized treatment. AI models have successfully predicted gene status related to OC prognosis by analyzing texture features extracted from CT images, with no significant correlation with BRCA mutation status, possibly due to the small number of patients evaluated. Additionally, AI models can predict Ki-67 status in OC through PET/CT image analysis, potentially becoming a new marker to replace Ki-67 detection.

Predicting Drug Treatment Efficacy and Prognosis

In drug efficacy and prognosis prediction, AI technology predicts OC patients' response to specific drugs by analyzing pathological images and clinical data. For example, deep learning models based on whole-slide imaging (WSI) can effectively distinguish OC patients' different responses to platinum-based drugs, showing high sensitivity and specificity. Additionally, AI models can predict the treatment effects of drugs like bevacizumab, providing a powerful tool for personalized OC treatment.

Multi-Omics Data Integration in Ovarian Cancer Research

Besides radiomics, AI technology integrates multi-omics data, including genomics, transcriptomics, proteomics, and metabolomics, to explore disease occurrence and development.

In genomics, AI analyzes multi-omics OC data, identifying OC subtypes by integrating different datasets and using denoising autoencoder frameworks. Studies show this method effectively identifies OC subtypes at the molecular level, screening target genes and KEGG pathways related to specific molecular subtypes. Additionally, AI models use methylation analysis of circulating tumor DNA (cfDNA) for OC diagnosis, showing high accuracy.

In transcriptomics, AI screens microRNAs (miRNAs) related to OC occurrence, which may be involved in cancer cell epithelial-mesenchymal transition and tumor heterogeneity and adaptability. Machine learning models based on serum miRNA profiles show good diagnostic performance for OC.

In metabolomics, AI analyzes metabolites in serum samples, screening OC-related metabolites and significantly improving early OC diagnosis performance by combining risk algorithms.

Challenges and limitations of AI in ovarian cancer care

Ovarian cancer is a formidable foe, with its insidious nature and often late detection posing significant challenges in diagnosis and treatment. Conventional methods have struggled to keep pace with the disease's complexity, leaving patients and healthcare providers grappling with limited options and uncertain outcomes. However, the advent of Artificial Intelligence (AI) has ushered in a new era of hope, offering unprecedented opportunities to revolutionize ovarian cancer care.

Despite its immense potential, the integration of AI into ovarian cancer care is not without hurdles. One of the primary obstacles lies in the scarcity of high-quality, diverse data sets required to train AI algorithms effectively. Ovarian cancer is a heterogeneous disease, with variations in tumor characteristics, patient demographics, and treatment responses, making it challenging to develop universally applicable AI models.

Furthermore, the interpretability and transparency of AI algorithms remain a concern. Healthcare professionals and patients alike demand a clear understanding of the decision-making processes underlying AI-driven recommendations, particularly in high-stakes scenarios like cancer treatment. Ensuring that AI systems are explainable and trustworthy is crucial for their widespread adoption in clinical settings.

Conclusion and Future Prospects

AI has shown satisfactory results in OC diagnosis, differentiation, and prognosis prediction. Combining AI models with traditional clinical diagnosis can improve diagnostic accuracy and efficiency, potentially enhancing future diagnostic systems. Additionally, predicting pathological subtypes and gene status may become a "virtual biopsy," reducing the need for invasive tests in the future.

Despite AI's potential in OC management, its clinical application faces challenges, including insufficient multi-omics data integration and single AI algorithms. Future research needs multi-center, large-sample validation to enhance AI models' universality and effectiveness and continuously optimize algorithms. Currently, AI models are mainly applied to thyroid, breast, and liver diseases, with OC applications still in early stages. AI's further development will significantly impact precision medicine, potentially playing a key role in OC treatment and management.

As AI continues to evolve and overcome these hurdles, the future of ovarian cancer care will be marked by unprecedented precision, personalization, and improved outcomes for patients worldwide. By harnessing the power of AI, we can unlock new frontiers in the fight against ovarian cancer, offering hope and empowering patients with the best possible care.

References

[1] Wang Y, Lin W, Zhuang X, Wang X, He Y, Li L, Lyu G. Advances in artificial intelligence for the diagnosis and treatment of ovarian cancer (Review). Oncol Rep. 2024 Mar;51(3):46.
[2] Allemani C, Weir HK, et al. Global surveillance of cancer survival 1995–2009: Analysis of individual data for 25,676,887 patients from 279 population-based registries in 67 countries (CONCORD-2) Lancet. 2015;385:977–1010.