A new deep learning framework can accurately classify four molecular subgroups of medulloblastoma and predict critical genetic risk factors using magnetic resonance imaging, according to a study by researchers from China. The artificial intelligence model achieved a median accuracy of 77.5% for subgroup classification and up to 91.3% for predicting high-risk genetic alterations. This approach could help clinicians stratify risk and tailor therapies without invasive testing.

♦Successful development of a stealth cloak (invisibility cloak) to protect nanomachines introduced into living organisms from foreign body reactions over extended periods: Constructed a stable ion pair network composed of polyanions and polycations on the nanomachine surface, devising a structure that prevents protein adsorption and attacks from macrophages.

♦Achieved ultra-long circulation in vivo with a half-life exceeding 100 hours after intravenous administration (10 times longer than previous stealth DDS systems).

♦Nanomachines equipped with asparaginase, an enzyme that breaks down L-asparagine, circulate long-term within the body, depleting L-asparagine—essential for cancer cell survival—from tumor tissues.

♦Confirmed the efficacy of starvation therapy for refractory breast cancer using a mouse model.

♦Furthermore, it breaks down the thick stroma (fibrous tissue) that blocks drug penetration into pancreatic cancer (such as immune checkpoint inhibitors), achieving extremely high efficacy (long-term survival) through synergistic effects with cancer immunotherapy.

♦A paper detailing the presentation content：

Junjie Li^*, Kazuko Toh, Panyue Wen, Xueying Liu, Anjaneyulu Dirisala, Haochen Guo, Joachim F. R. Van Guyse, Saed Abbasi, Yasutaka Anraku, Yuki Mochida, Hiroaki Kinoh, Horacio Cabral, Masaru Tanaka, and Kazunori Kataoka^*

Nature Biomedical Engineering (2025)

https://doi.org/10.1038/s41551-025-01534-1