Artificial intelligence (AI) has reached a new milestone in the field of oncology, with the creation of an advanced model developed by scientists at Harvard Medical School. This system, described on September 4, 2024 in the journal Nature, has been Designed to perform diagnostic and prediction tasks in 19 different types of cancer, providing medical professionals with an unprecedented multifunctional tool.
CHIEF has been designed to perform diagnostic and prediction tasks in 19 different types of cancer, providing an unprecedented multifunctional tool.
The model, called CHIEF (Clinical Histopathology Imaging Evaluation Foundation)offers flexibility and versatility similar to ChatGPT, but geared towards the medical field. Its ability to multitask and adapt to different cancer types sets it apart from current AI tools, which tend to be limited to a small number of cancers or specific diagnostic tasks.
AI in oncology
Traditional artificial intelligence systems used in cancer diagnosis have focused on specific tasks, such as detecting cancer cells or predicting tumor genetic profiles. These approaches are useful, but have limitations, as they are often effective only in specific types of cancer and in specific clinical contexts. In contrast, CHIEF has been trained to tackle a broader spectrum of tasks, from tumor detection to predicting treatment response and assessing patient prognosis.
CHIEF has been trained to tackle a broader spectrum of tasks, from tumor detection to predicting response to treatments and assessing patient prognosis.
In this sense, Kun-Hsing Yu, senior author of the study and adjunct professor of biomedical informatics at the Blavatnik Institute at Harvard Medical Schoolhe explained to the media worldpharmanews The goal was to create an agile and versatile artificial intelligence platform, similar to ChatGPT, that could perform a wide range of tasks related to cancer assessment.
Comprehensive approach
CHIEF works by analyzing digital images of tumor tissue. From these images, it is able to detect cancer cells with greater accuracy than many AI systems currently available. In addition, predicts the molecular profile of the tumor by observing specific cellular characteristics present in the imagesThis approach makes it possible, for example, to predict patient survival and to identify the tumor microenvironment, i.e. the characteristics of the tissue surrounding the tumor that may influence the patient’s response to treatments such as chemotherapy, radiotherapy and immunotherapy.
CHIEF predicts the molecular profile of the tumor by observing specific cellular characteristics present in the images
This model has a significant impact on physicians’ ability to assess cancers more quickly and accurately. One of its most notable features is its ability to identify patients who may not respond adequately to standard cancer therapies, which could facilitate the selection of more effective experimental treatments.
I noted that if it is validated and implemented on a large scale, CHIEF and similar approaches could identify patients with specific molecular variations early, improving the personalization of treatments.This is particularly valuable in resource-limited clinical settings, where genetic sequencing is not always available.
Training and validation
The CHIEF model was initially trained on 15 million unlabeled images, and its skills were subsequently refined on 60,000 whole tissue images of common cancers, including lung, breast, prostate, colon, and stomach.This approach allowed CHIEF to develop a deep understanding of not only specific areas within images, but also the overall context of each tumor. As a result, the system can interpret images more holistically, which sets it apart from other AI models that focus on specific regions of tissues.
Validation was carried out using more than 19,400 tumor slide images from 24 hospitals and patient cohorts worldwide.
CHIEF validation was performed using over 19,400 tumor slide images from 24 hospitals and patient cohorts worldwide.The results were impressive: CHIEF outperformed other AI systems by up to 36 percent on key tasks such as detecting cancer cells, predicting clinical outcomes, and identifying genetic patterns related to response to treatments.
One of the most notable advantages of CHIEF is its ability to work equally well regardless of how tumor samples were obtained (by biopsy or surgical excision) or how they were digitized. This means that the model can be adapted to different clinical environments and sampling techniques, increasing its potential for use in hospitals around the world..
Genomic precision and prediction
Not only does CHIEF excel at cancer detection, achieving 94 percent accuracy across multiple cancer types, it is also able to successfully predict key genetic variations in tumors. These genomic predictions, based solely on pathological images, offer a rapid and cost-effective alternative to traditional genetic sequencing. DNA sequencing, while essential in many cases, can be expensive and slow, with wait times of weeks for results. Instead, CHIEF AI can identify genetic patterns in minutes, accelerating therapeutic decision-making.
These genomic predictions, based solely on pathological images, offer a rapid and cost-effective alternative to traditional genetic sequencing.
In a particularly revealing experiment, CHIEF predicted key mutations associated with response to FDA-approved treatments in 18 genes linked to 15 types of cancer. This included a 96 percent accuracy in detecting mutations in the EZH2 gene in large B-cell lymphomas, a common type of blood cancer.
CHIEF’s Future
The researchers behind CHIEF are already working to expand its capabilities, with the goal of including imaging of rare diseases and premalignant tissue. They also plan to train the model to predict not only the response to standard treatments, but also the benefits and adverse effects of new cancer therapies.
This AI system promises to transform the way cancer is addressed in the coming years
The breakthrough achieved by CHIEF marks an important step in the fight against cancer, providing physicians with a tool that not only improves diagnostic accuracy but also accelerates the identification of personalized treatment options. While further studies are still needed to perfect its performance, this AI system promises to transform the way cancer is approached in the coming years, offering hope to millions of patients around the world.
