TNM Staging

The T category describes the size and local extent of the primary tumor. AI models analyze tissue architecture and cellular morphology to classify the depth of invasion and tumor dimensions.

• TX: Tumor cannot be assessed.
• T0: No evidence of primary tumor.
• Tis: Carcinoma in situ (pre-invasive).
• T1-T4: Increasing size and/or local extension.

For example, a model might distinguish T1 (tumor ≤2cm) from T2 (tumor >2cm but ≤5cm) by learning spatial features from whole slide images.

The N category indicates the absence or presence and extent of regional lymph node metastasis. Computational pathology models detect micrometastases and isolated tumor cells that may be missed by manual examination.

• NX: Regional lymph nodes cannot be assessed.
• N0: No regional lymph node metastasis.
• N1-N3: Increasing number, size, or location of involved nodes.

AI-driven lymph node segmentation and metastasis detection algorithms quantify tumor burden within node tissue, providing objective N-staging inputs.

The M category specifies whether the cancer has spread to distant organs or tissues. While often determined by radiology, pathological confirmation via biopsy is definitive.

• M0: No distant metastasis.
• M1: Distant metastasis present.

AI models can identify metastatic carcinoma in biopsies from distant sites and, when combined with multi-modal diagnostic fusion, correlate pathological findings with imaging evidence to confirm M1 status.

Stage grouping combines T, N, and M categories into an overall prognostic stage (Stage 0 through Stage IV). This synthesis drives treatment decisions and clinical trial eligibility.

• Stage 0: Carcinoma in situ (Tis, N0, M0).
• Stage I: Early-stage, localized disease.
• Stage II-III: Locally advanced disease.
• Stage IV: Metastatic disease (any T, any N, M1).

AI systems aggregate individual T, N, and M predictions into a final stage group, often using rule-based post-processing or learned ordinal regression to ensure clinical consistency.

While not part of the TNM acronym, histological grade is a critical complementary component. It quantifies how much tumor cells resemble normal tissue under the microscope.

• GX: Grade cannot be assessed.
• G1: Well-differentiated (low grade).
• G2: Moderately differentiated.
• G3: Poorly differentiated (high grade).
• G4: Undifferentiated.

Deep learning models like those for Gleason grading in prostate cancer automate this assessment by analyzing glandular architecture and nuclear atypia, providing a reproducible grade that refines the TNM-based prognosis.

After neoadjuvant therapy, the R classification and ypTNM prefix denote residual tumor status and post-treatment staging. AI models assess treatment response by quantifying viable tumor cells versus fibrosis.

• R0: No residual tumor (complete response).
• R1: Microscopic residual tumor.
• R2: Macroscopic residual tumor.

Computational tumor-stroma ratio analysis and tumor-infiltrating lymphocyte (TIL) quantification provide granular metrics of pathological complete response, informing adjuvant therapy decisions.

A weakly supervised learning paradigm essential for slide-level TNM classification. In MIL, a WSI is treated as a bag of patches (instances), and only the slide-level stage label is provided during training. The model learns to identify which patches contain diagnostically relevant features—such as tumor morphology or lymphovascular invasion—without requiring pixel-level annotations. Architectures like CLAM and attention-based MIL pooling are the standard for this task.

Immune cells that have migrated into the tumor microenvironment, quantified as a prognostic biomarker directly from H&E-stained slides. In the context of TNM staging, stromal TIL density correlates with lymph node metastasis risk and overall survival, particularly in triple-negative breast cancer and melanoma. AI models can automatically segment and enumerate TILs, providing a quantitative immune score that complements traditional anatomical staging.

The presence of tumor cells within lymphatic channels or blood vessels is a critical histologic feature that upstages a tumor in the TNM system. AI models are trained to detect these subtle morphological patterns, which appear as tumor emboli within endothelial-lined spaces. Automated detection reduces inter-pathologist variability and flags high-risk cases where micrometastases may be present despite negative nodal imaging.

A deep learning approach that models the tumor microenvironment as a graph, where nodes represent tissue patches and edges represent spatial relationships. This explicitly captures the architectural patterns relevant to TNM staging—such as the distance between invasive tumor nests and lymphovascular structures. GNNs can learn that a small tumor with extensive perineural invasion may behave more aggressively than its T-stage suggests.