Researchers have just developed an artificial intelligence capable of finding the origin of metastatic tumors while generating a “differential diagnosis” for patients with cancers of unknown origin.
- Based on deep learning, this artificial intelligence sifted through more than 22,000 cancer images to precisely determine the origin of metastatic tumors.
- The AI correctly identified the cancer in 83% when the latter was of known origin, and with an accuracy of 61% in the case of an unknown primary cancer.
For cancer patients, knowing the primary site of origin of the tumor is essential to allow targeted action and thus increase the survival rate. However, even today, it remains impossible to determine the origin of metastatic tumors in 1 to 2% of cancer cases. However, the prognosis of an unknown primary cancer (IPC) is poor, with a median overall survival of 2.7 to 16 months.
In order to receive a more accurate diagnosis, patients often must undergo extensive and invasive diagnostic tests, which can delay treatment. But artificial intelligence (AI) could improve diagnosis for patients with complex metastatic cancers. Developed by researchers at the Mahmood Laboratory at Brigham and Women’s Hospital, it uses routinely acquired histology slides to precisely find the origins of metastatic tumors while generating a “differential diagnosis”, for patients with CUP. . The results have just been published in the journal Nature.
“Almost all patients diagnosed with cancer have a histology slide, which has been the diagnostic standard for over a hundred years. Our work provides a way to leverage universally acquired data and the power of intelligence to improve the diagnosis of these complicated cases which generally require in-depth diagnostic work”explains Faisal Mahmood, who led the work.
83% accuracy for cancers of known origin
Called Tumor Origin Assessment via Deep Learning (TOAD), this AI is a deep learning-based algorithm that simultaneously identifies the tumor as primary or metastatic and predicts its site of origin. To train their model and make it more accurate, the researchers had it analyze pathological images of tumors from more than 22,000 cancer cases. They then tested TOAD on approximately 6,500 cases of known primary origin and analyzed increasingly complicated metastatic cancers to establish the utility of the AI model on CUPs.
For tumors whose primary origin was known, the model correctly identified the cancer in 83% of cases and ranked the diagnosis among its top three predictions in 96% of cases.
61% accuracy for unknown primary cancers
The researchers then tested the model on 317 cases of cervical cancer for which a differential diagnosis had been assigned. They found that the diagnosis of TOAD matched pathologist reports in 61% of cases and was in the top three in 82% of cases.
Now, the researchers hope to continue training their histology-based model with more cases and engage in clinical trials to determine if it improves diagnostic capabilities and patient prognosis.
“The model’s superior predictions can speed up diagnosis and subsequent treatment by reducing the number of ancillary tests to be ordered, reducing the removal of additional tissue, and the overall time needed to diagnose patients, which can be time-consuming and stressful”, says Professor Mahmood. The researcher believes that it is only a question here of “the first step in using whole slide images for AI-assisted cancer origin prediction, and this is a very exciting area that has the potential to standardize and improve the diagnostic process “.
.
