AI in Research Institution

The AI-IR lab at the University of Texas Medical Branch reimplemented 21 pretrained medical imaging models, including nnU-Net, TotalSegmentator, MedSAM2 and SAM 3, natively in MLX for Apple Silicon. Benchmarked against the original implementations on an Intel CPU and an NVIDIA RTX 4090 across public cases from 11 imaging datasets, the models matched outputs in 211 of 212 comparisons, and 19 of the 21 ran in under 16 GB of memory on a 64 GB Apple M1 Max, with a memory-saving setting cutting memory use by a median 49% at a median 13% processing-time cost.

Johns Hopkins researchers released Merlin-Cancer-Super, a model that detects and segments tumors in nine organs on CT scans. It is trained with R-Super, which uses an LLM to extract tumor count, size and location from radiology reports and turns them into per-voxel training signals, on Merlin Plus, a dataset built with Stanford from 25,494 CT scans. The authors report it outperforms leading public models and vision-language models.

Zhejiang Cancer Hospital and Alibaba DAMO Academy developed DAMO GRAPE, the first non-contrast CT AI for gastric cancer screening, achieving 85.1% sensitivity and 96.8% specificity, with retrospective studies showing earlier detection of 11 cases by 2-10 months. The model is being deployed in Zhejia

Creval Sistemi e Servizi, internal IT unit of Italian banking group Credito Valtellinese (Creval), deployed IBM Watson-powered virtual assistant Alfredo across 360+ bank branches, reducing back-office service desk call volume by 80%.

Lawrence Berkeley National Laboratory's A-Lab combines literature-trained ML models, an active-learning algorithm (ARROWS3), ML analysis of X-ray diffraction and robotic synthesis. Over 17 days of continuous operation it obtained 36 of 57 target oxides and phosphates selected from Materials Project and Google DeepMind stability data, per the Nature paper, which received an author correction in January 2026.

China Agricultural University built Shennong, a vertical large language model for agriculture, from 3,000+ scanned books, field data from 20+ provinces and a 3 PB knowledge base. Version 3.0 (October 2025) adopted a lightweight multi-agent design that halved compute use and added 36 specialised agents. It serves over 100,000 farmers in pilot applications and covers 90% of agricultural disciplines.

Researchers at the Advanced Research and Innovation Center of Khalifa University in Abu Dhabi propose a robotic system that localizes, picks, images and inspects aeroengine blades using two trained deep learning models. In controlled-condition experiments the defect model reached an mAP of 88.2% and a blade took approximately 4 seconds per inspection cycle.

After upgrading its rule-based EduBot chatbot with AI on 28 May 2025, the University of Cape Town reports 245,000 AI-powered answers, a 4.8/5 rating from 3,823 user ratings, and 93% of users answered without human intervention. Requests for human assistance fell from 11% to 8%, per a Verge AI release.