In an exciting collaboration between MIT and Dana-Farber Cancer Institute, researchers have developed a revolutionary AI model that could help doctors identify the origin of certain cancers. Named OncoNPC, the model is designed to guide treatment decisions for cancers of unknown primary (CUP), thereby improving outcomes and broadening treatment options.
Identifying Cancer's Mysterious Origins
Determining the origin of a patient's cancer is crucial for selecting targeted treatments. However, in 3-5% of cases, particularly when the cancer has spread throughout the body, oncologists find it challenging to determine where the cancer began. These tumors, classified as CUP, often limit the treatment options, as precision drugs are typically designed for specific types of cancer.
OncoNPC: The Game Changer
OncoNPC, short for Oncology NGS-based Primary cancer type Classifier, is an AI-based tool trained on genetic sequences from nearly 30,000 patients diagnosed with one of 22 known cancer types. The model was developed using a machine learning technique that offers more transparency in its reasoning, aiding clinicians' trust in the model.
Using a set of about 400 genes often mutated in cancer, OncoNPC analyzes the sequence and predicts the tumor's origin with about 80% accuracy. For high-confidence predictions, the accuracy rises to roughly 95%.
Advancing Treatment Options
The OncoNPC model could be transformative for patient care. In a dataset of about 900 patients, researchers discovered that the model could classify at least 40% of tumors of unknown origin with high confidence, a 2.2-fold increase in the number of patients who could be eligible for genomically guided targeted treatment.
Among CUP patients who received a targeted treatment, those who received a treatment aligned with the type of cancer predicted by the model fared better. The researchers also identified an additional 15% of patients who could have benefited from existing precision treatments.
This has the potential to broaden treatment avenues for a significant number of patients diagnosed with CUP, without requiring new drugs to be approved.
Future Prospects
While the model's prediction has been validated using retrospective data only, it may be further tested in clinical trials. The team also plans to expand the data used for prediction to include additional diagnostic information, such as pathology results and radiology images. This comprehensive perspective could enable the model to predict not only the type of tumor and patient outcome but also the optimal treatment.
The researchers are also considering collaboration with community cancer centers, particularly where cases of CUP might be more common due to limited resources.
Conclusion
The OncoNPC model represents a significant stride in cancer diagnosis and treatment. By accurately predicting the origin of elusive cancers, it opens the door to more precise and effective treatments, potentially transforming the care of patients with CUP.
The tool's ease of use and wide applicability, especially in settings with limited resources, underline its potential as a valuable addition to the arsenal in the fight against cancer. The research, published in Nature Medicine, highlights the promising future of AI in advancing medical science and personalized patient care.
Identifying Cancer's Mysterious Origins
Determining the origin of a patient's cancer is crucial for selecting targeted treatments. However, in 3-5% of cases, particularly when the cancer has spread throughout the body, oncologists find it challenging to determine where the cancer began. These tumors, classified as CUP, often limit the treatment options, as precision drugs are typically designed for specific types of cancer.
OncoNPC: The Game Changer
OncoNPC, short for Oncology NGS-based Primary cancer type Classifier, is an AI-based tool trained on genetic sequences from nearly 30,000 patients diagnosed with one of 22 known cancer types. The model was developed using a machine learning technique that offers more transparency in its reasoning, aiding clinicians' trust in the model.
Using a set of about 400 genes often mutated in cancer, OncoNPC analyzes the sequence and predicts the tumor's origin with about 80% accuracy. For high-confidence predictions, the accuracy rises to roughly 95%.
Advancing Treatment Options
The OncoNPC model could be transformative for patient care. In a dataset of about 900 patients, researchers discovered that the model could classify at least 40% of tumors of unknown origin with high confidence, a 2.2-fold increase in the number of patients who could be eligible for genomically guided targeted treatment.
Among CUP patients who received a targeted treatment, those who received a treatment aligned with the type of cancer predicted by the model fared better. The researchers also identified an additional 15% of patients who could have benefited from existing precision treatments.
This has the potential to broaden treatment avenues for a significant number of patients diagnosed with CUP, without requiring new drugs to be approved.
Future Prospects
While the model's prediction has been validated using retrospective data only, it may be further tested in clinical trials. The team also plans to expand the data used for prediction to include additional diagnostic information, such as pathology results and radiology images. This comprehensive perspective could enable the model to predict not only the type of tumor and patient outcome but also the optimal treatment.
The researchers are also considering collaboration with community cancer centers, particularly where cases of CUP might be more common due to limited resources.
Conclusion
The OncoNPC model represents a significant stride in cancer diagnosis and treatment. By accurately predicting the origin of elusive cancers, it opens the door to more precise and effective treatments, potentially transforming the care of patients with CUP.
The tool's ease of use and wide applicability, especially in settings with limited resources, underline its potential as a valuable addition to the arsenal in the fight against cancer. The research, published in Nature Medicine, highlights the promising future of AI in advancing medical science and personalized patient care.