A new AI tool from Harvard Medical School can accurately detect and analyze cancer cells.

The revolutionary new model is called the Clinical Histopathology Imaging Evaluation Foundation and known as ‘CHIEF’.

The model operates similarly to a ChatGPT chatbot and can diagnose cancer, guide treatment choice and predict patient outcomes. CHIEF's design comes from a team of scientists at Harvard Medical School. Their innovative approach leverages the AI algorithms to analyze medical images and provide more accurate and efficient diagnoses

The model was 36% better than other leading AI methods in cancer cell detection, tumor origin identification, predicting patient outcomes, and identifying the presence of genes and DNA patterns related to how patients respond to treatments.

em360tech image

Read: Google's Med-Gemini AI is here. Will it Revolutionize Healthcare?

Typically, AI systems are trained to perform specified tasks such as detecting cancer or predicting a genetic profile and only work on specific cancer types. However, this new model can perform these tasks simultaneously and was tested on 19 different types of cancer.

CHIEF was initially published in Nature in September 2024. It remains a research model rather than a cleared clinical tool, but the field around it has moved quickly, and the Harvard team has since turned to making pathology AI fairer.

How Does the CHIEF AI Model Work?

The chief AI model works by reading digital slides of tumor tissues. From there it is able to detect cancer cells and predict the molecular profile of the tumor based on the cellular features.

Through this, it is able to forecast patient outcomes across multiple cancer types and accurately identify features around the tumor microenvironment. These can influence a patient's response to treatments including surgery, chemotherapy, immunotherapy and radiation.

Read: What Is Health Technology

The team also reports that CHIEF AI appears to be able to generate novel insights including identifying specific tumor characteristics not previously thought to be related to patient outcomes.

The CHIEF AI tool was initially trained on 15 million unlabeled images before being trained on 60,000 whole-slide images of tissues of multiple forms of cancer. By training the model to look at both the whole image and specific sections of an image it learns to relate changes in one region to the overall context. This enabled a more holistic interpretation of the images.

The model was then tested on 19,400 whole-slide images from 32 independent datasets originating from 24 different hospitals across the world.

Pathology is the study of tissue samples under a microscope. It’s how cancers are confirmed, classified and graded, and CHIEF is trained on digitised versions of those slides.

How Will CHIEF AI Be Used?

The results of CHIEF AI testing add to the evidence that an AI-powered revolution is coming to healthcare, enhancing clinicians' abilities to identify and evaluate cancer as well as being able to tell earlier on if specific patients may not respond well to certain therapies.

In testing, CHIEF detected cancer with nearly 94% accuracy across 15 datasets covering 11 cancer types, reached 96% on biopsy datasets, and predicted mutations across 54 commonly mutated cancer genes with over 70% overall accuracy.

Can Pathology AI be Biased?

In December 2025, Kun-Hsing Yu, who led the CHIEF study, and his team reported in Cell Reports Medicine that four widely used pathology AI models performed unevenly across patients’ race, gender and age, with gaps in roughly 29% of the diagnostic tasks analysed. The researchers found the models can pick up demographic signals from tissue slides and use them as shortcuts. Their fix, a training framework called FAIR-Path, cut the disparities by about 88%.

What is Next for CHIEF AI?

When the study was published, the Harvard Medical School team planneds to further refine the model's performance and augment its capabilities further.

This includes conducting additional training that will mean the model can be used for rare diseases and non-cancer issues. This expanded training will also include more samples from malignant tissues before cells become fully cancerous.

By feeding the model more molecular data it will be better able to identify cancers with different levels of aggressiveness. The model will also be intentionally trained to predict the benefits and drawbacks of new cancer treatments alongside other standard treatments.

Pathology AI is already reaching clinics. In August 2025 the FDA granted de novo authorisation to ArteraAI Prostate, the first AI digital pathology tool cleared to give both prognostic and predictive information in localised prostate cancer. In May 2026 ArteraAI became the first FDA-cleared digital pathology risk stratification tool for cancer. Those are narrower tools than a general-purpose model like CHIEF, and regulators are still working out how to handle broad foundation models. As one industry review put it, regulatory clearance is only one step towards clinical adoption.