Chinese researchers have developed an algorithm capable of determining the origin of metastases from 5 classes of cancer (stomach, breast, ovaries, lung, blood) even though the tumor of origin was unknown. The results were published in Nature Medicine [1].
The researchers showed the algorithm 30,000 images of metastatic cells from the abdominal or pleural fluid of 21,000 patients whose origin of the tumor was known in order to “ train “. The algorithm learned to associate the morphological characteristics of cancer cells according to their hematological, gynecological, digestive or pulmonary origin. The researchers then asked artificial intelligence (AI) to find the origin of the metastases on its own using 27,000 new images. In 80% of cases the algorithm succeeded in identifying the organ initially affected.
Improved survival times
To determine the clinical benefit of AI, the researchers retrospectively evaluated the impact on survival of the treatment that would have been given, based on the origin of the cancer predicted by the algorithm for 391 patients treated during the four years preceding the study. Half of the patients whose initial treatment protocol was consistent with the origin of metastases predicted by the AI were alive twenty-seven months later, compared to seventeen months for those who had received discordant treatment.
“ The choice of cancer treatment (…) is normally guided by the characteristics of the initial tumor. However, patients for whom the origin of metastases is unknown are generally treated using “broad spectrum” chemotherapies which are not specific to their disease. » explains Professor Arnaud Scherpereel, pneumo-oncologist at Lille University Hospital. The benefit of these chemotherapies is less than 20%, and the overall survival of patients is on average less than one year from the announcement of the diagnosis. “ In the future, AI could allow more patients to know the initial site of the cancer and therefore directly apply the most suitable treatment. » says Sarah Watson, researcher at the Institut Curie.
An algorithm based on cytological data
“ The novelty of this algorithm is that it is based on cytological data, therefore liquid samples of malignant tissues, while the other AIs were trained on biopsies or even on molecular data from tumor sequencing. » further emphasizes Sarah Watson.
However, AI cannot distinguish between tumor subtypes. “ AI predictions already make it possible to eliminate ineffective treatments, for example by rejecting the idea of using a treatment against breast cancer when the metastases are of pulmonary origin. » notes Professor Arnaud Scherpereel. “ But doubts persist about possible treatments among the different respiratory cancers that are more or less sensitive to chemotherapy, targeted therapies or immunotherapy. » he specifies.
Cancers of unknown primary (CPI) affect 7,000 patients in France each year.
[1] Tian, F., Liu, D., Wei, N. et al. Tumor origin prediction in cancers of unknown primary origin using cytology-based deep learning. Nat Med (2024). https://doi.org/10.1038/s41591-024-02915-w
Source: Le Figaro, Elisa Doré (04/22/2034)
