How AI in Oncology is Changing the Cancer Fight (Part 2)

Introduction

AI is changing cancer care by improving early detection, diagnosis, treatment, and drug discovery. This article explains key performance measures like AUC and Dice scores, then highlights real-world research showing AI’s impact. Following our previous discussion on AI technologies in oncology, we now focus on practical results shaping cancer treatment.

How AI’s Performance is Measured in Cancer Care

To assess AI’s effectiveness in oncology, key metrics evaluate accuracy, reliability, and adaptability.

Evaluating Model Performance

• AUC/AUROC: Measures how well the model distinguishes between cancerous and non-cancerous cases. Higher scores indicate better performance.
• Accuracy: Shows the percentage of correct predictions but may be misleading with imbalanced data.
• Dice Score: Assesses how closely AI-detected tumor areas match actual ones.

Diagnostic Reliability

• Sensitivity: Measures how well AI identifies cancer cases, reducing missed diagnoses.
• Specificity: Ensures AI avoids false positives, preventing unnecessary treatments.

Validation for Real-World Use

• Internal Validation: Tests AI with familiar data in controlled settings.
• External Validation: Confirms AI works effectively with new data from different sources.

These methods ensure AI in cancer care is accurate, reliable, and adaptable for clinical applications.

AI’s Role Across the Oncology Industries

This article features hand-picked research on how AI enhances early cancer detection, diagnosis, and personalized treatment using technologies like deep learning, radiomics, and predictive analytics.

Early Detection: Spotting Cancer Before It Strikes

AI is advancing early cancer detection with innovative, non-invasive methods.

• Predicting High-Risk Patients: A 2023 Nature Medicine study introduced CancerRiskNet, an AI model analyzing over 9 million records to predict high-risk pancreatic cancer cases up to three years earlier (AUROC 0.88 in Denmark, 0.78 in the US after retraining).

• Cancer Diagnosing with Liquid Biopsies: Shivashankar et al. (2024) developed an AI tool analyzing chromatin in blood samples, achieving 77% accuracy in detecting cancer and 78% in classifying tumor types. It also monitored treatment effects, offering insights into tumor behavior.

Diagnostic: Enhancing Accuracy and Efficiency

AI is improving accuracy and efficiency in cancer diagnostics through medical imaging and digital pathology.

• Bladder Cancer Detection: FGP-Net (Zhang et al., 2021) analyzed CT scans to assess muscle invasion, achieving AUC scores of 0.861 (internal) and 0.791 (external), outperforming radiologists.

• Breast Cancer Diagnosis: Jiménez Gaona et al. (2024) used convolutional neural networks (CNNs) and generative adversarial networks (GANs) to enhance overcome limited data and to generate synthetic medical images, achieving an AUC of 0.88 and improving lesion detection.
• Prostate Cancer Assessment: Tolkach et al. (2023) developed an AI tool analyzing 5,900 biopsy slides, achieving 97.1–100% sensitivity and performing on par with expert pathologists.

Precision Oncology Treatment with AI

AI is revolutionizing cancer treatment by personalizing therapies based on genetic and clinical data.

• Targeted Therapies: Chen et al. (2020) developed an AI model using H&E-stained images to differentiate benign from malignant liver tissue with 96% accuracy, aiding precise diagnosis and treatment.
• Immunotherapy & Survival Prediction: Chen et al. (2024) created an AI tool to detect tertiary lymphoid structures (TLSs) in tissue samples, achieving Dice scores of 0.91 (internal) and 0.866 (external). Higher TLS levels correlated with better immune response and improved survival predictions in 10 of 15 cancer types.

Predictive Analytics: Guiding Therapy Decisions

AI predicts treatment responses, monitors tumor changes in real time, and adapts therapies for better outcomes.

Predicting Treatment Responses:

• Melanoma Metastases: Salgado & AbdulJabbar (2023) developed an eTIL scoring system to assess tumor-infiltrating lymphocytes, predicting metastasis risk and response to anti-PD-1 immunotherapy.
• Breast Cancer Chemotherapy: Zhi Huang (Nature Cancer) created an AI model (IMPRESS pipeline) to predict patient responses to neoadjuvant chemotherapy, achieving AUCs of 0.8975 (HER2-positive) and 0.7674 (triple-negative breast cancer).

Real-Time Tumor Monitoring

• Nasopharyngeal Cancer: AI models (HoverNet, MorphResNet) analyzed 385 samples, using TIL scores to predict recurrence risk with 92.1% accuracy.
• Adaptive Breast Cancer Radiotherapy: Feng et al. (2023) tested an AI segmentation tool for CBCT images, achieving 98% accuracy in identifying key structures, ensuring precise treatment adjustments.

Accelerating Drug Discovery with AI

AI accelerates cancer drug development and optimizes clinical trials by analyzing vast datasets in seconds.

Faster Drug Discovery

• Pinpointing Drug Targets: Abel et al. (2024) developed an AI model (Mask-RCNN) to analyze H&E-stained tissue slides, achieving a Dice score of 0.818. It identified links between cell nuclei characteristics and genetic mutations driving cancer progression.

Enhancing Clinical Trials:

• Predicting Patient Outcomes: Zhang et al. (2024) used AI models (LightGBM, Logistic Regression) to predict breast cancer metastasis, achieving AUC scores up to 0.971, improving early detection and treatment planning.
• Streamlining Data Analysis: Kaczmarzyk et al. (2024) introduced WSInfer, an open-source tool integrated with QuPath, enabling AI-driven pathology analysis for breast, colorectal, and lung cancers.

Conclusion

AI is revolutionizing cancer care by enhancing diagnostics, predicting treatment responses, and accelerating drug discovery. Innovations like CancerRiskNet, WSInfer, and adaptive therapies are driving major advancements.

Challenges such as data quality and accessibility remain, but AI’s potential continues to grow. At SciForce, we develop AI-driven healthcare solutions to advance oncology.

For a more detailed research overview, read the full article on our website.