Teaching pack: AI & Computation

AI and computation collects the software that reads scans and slides, predicts outcomes, designs drugs and matches patients to trials. It covers FDA-cleared digital pathology risk tools such as ArteraAI, radiology triage and screening models, pathology and radiology foundation models, multimodal patient-level models, AI-driven target discovery and ADC design, and LLM-based trial matching and tumour-board support. Listed technologies include federated learning, CHIEF, Phikon, CT-FM, MedSAM, Aidoc CARE, CellFM, GenePT, Nucleotide Transformer, Enformer and Borzoi, and NVIDIA BioNeMo. Records pointing here include AI in radiology, digital pathology and AI, AI trial matching, AI-driven drug and target discovery, and the companies Tempus AI, BostonGene, Artera, Paige AI, PathAI and Owkin.

• Adaptive therapy (evolution-based dosing): Instead of hitting a tumour as hard as possible, adaptive therapy gives just enough drug to keep it in check and stops when it shrinks, so drug-sensitive cells survive to compete with resistant ones. A pilot trial in prostate cancer roughly doubled the time to progression on abiraterone.
• Agent-based and multicellular simulations: Instead of equations for average behaviour, agent-based models simulate every cell as an individual with rules for dividing, moving, dying and signalling, producing virtual tumours in which immune attack, drug delivery and evolution can be watched and tested.
• AI auto-contouring and adaptive planning: Software that draws organs and tumours on scans automatically, saving hours per patient and making daily plan adaptation practical.
• AI compute and model platforms for oncology: AI compute platforms are the GPUs, model libraries, and cloud services that pathology, radiology, and drug-design AI run on.
• AI in radiology: Software that reads scans alongside radiologists, catching cancers earlier and predicting who is at risk.
• AI polyp detection in colonoscopy: Real-time software that highlights polyps on the colonoscopy screen, helping doctors find more of the growths that could become bowel cancer.
• AI scoring of biomarkers and grade on pathology slides: Software that reads digitised biopsy slides to detect cancer, grade it, and score biomarkers such as HER2, PD-L1 and Ki-67 more consistently than the eye alone.
• AI trial matching & clinical decision support: Software, increasingly LLM-based, that reads a patient's record and finds trials or guideline options they qualify for.
• AI-assisted mammography screening: Software that reads screening mammograms alongside or before radiologists, catching more cancers and cutting the reading workload in large trials.
• AI-driven drug & target discovery: Using machine learning to pick targets, design molecules and antibodies, and predict which ADC will work.

• AI in oncology roadmap: pattern readers → foundation models → agents in the workflow · Computer-aided detection (historic)
• AI in oncology roadmap: pattern readers → foundation models → agents in the workflow · Deep learning reaches cleared devices (historic)
• AI in oncology roadmap: pattern readers → foundation models → agents in the workflow · Structure prediction and generative design (current)
• AI in oncology roadmap: pattern readers → foundation models → agents in the workflow · Foundation models and the first predictive tests (current)
• AI in oncology roadmap: pattern readers → foundation models → agents in the workflow · Language models enter the workflow (emerging)
• AI in oncology roadmap: pattern readers → foundation models → agents in the workflow · From prediction to prospective proof (emerging)
• AI in the oncology clinic: from narrow cleared tools to multimodal decision support · Narrow detection tools cleared (historic)
• AI in the oncology clinic: from narrow cleared tools to multimodal decision support · Screening at scale and risk models (current)
• AI in the oncology clinic: from narrow cleared tools to multimodal decision support · First predictive AI and foundation models in products (current)
• AI in the oncology clinic: from narrow cleared tools to multimodal decision support · Prospective evidence and regulatory frameworks (emerging)
• AI in the oncology clinic: from narrow cleared tools to multimodal decision support · Speculative: multimodal decision support as standard of care (speculative)
• Virtual cell roadmap: from bulk omics to a predictive model of a cancer cell · Atlases and bulk omics (historic)
• Virtual cell roadmap: from bulk omics to a predictive model of a cancer cell · Pooled perturbation screens meet single cells (historic)
• Virtual cell roadmap: from bulk omics to a predictive model of a cancer cell · First single-cell foundation models (current)
• Virtual cell roadmap: from bulk omics to a predictive model of a cancer cell · Data at scale and context-aware models (current)
• Virtual cell roadmap: from bulk omics to a predictive model of a cancer cell · Patient-derived contexts and spatial niches (emerging)
• Virtual cell roadmap: from bulk omics to a predictive model of a cancer cell · Speculative: in silico trials and digital twins (speculative)

• MASAI (Mammography Screening with Artificial Intelligence) (phase 3, n=105,934): Cancer detection rate (interim analysis): 6.1 per 1,000 screened vs 5.1 per 1,000 screened
• ScreenTrustCAD (observational study, n=55,581): Screen-detected breast cancers (primary comparison: one radiologist plus AI versus two radiologists): 261 cases vs 250 cases
• NCI-COG Pediatric MATCH (APEC1621) (platform trial, n=1,377): Actionable alteration detected (first 1,000 tumours): 31.5%
• NCI-MATCH (EAY131) (platform trial, n=5,954): Assigned to a treatment arm: 17.8%
• I-PREDICT (phase 2): Consented patients who received molecularly matched personalised treatment: 49%
• WINTHER (phase 2, n=303): Progression-free survival ratio above 1.5 against the prior line of therapy: 22.4%

1. What is the first FDA-cleared AI pathology tool that predicts treatment benefit, and in which cancer?
   Answer

   ArteraAI Prostate (de novo authorisation, August 2025) predicts prognosis and benefit from short-term androgen deprivation with radiotherapy in localised prostate cancer.
2. What are foundation models in pathology and what have they enabled?
   Answer

   Very large self-supervised models trained on millions of slides (Virchow, UNI, CONCH, Prov-GigaPath) that adapt to many tasks; they predict molecular status from H&E and power FDA-cleared tools such as ArteraAI Prostate and ArteraAI Breast.

• NCI: Artificial intelligence in cancer research: https://www.cancer.gov/research/infrastructure/artificial-intelligence