aetherAI, Asia’s leading medical image AI solution provider focused on digital pathology, announced that the results of its method for training neural networks on entire (WSIs) using only slide-level diagnoses has been published in the peer-reviewed Nature Communications (Impact IF:14.92). The method leverages the unified memory mechanism to overcome the memory constraints of computer accelerators and demonstrates higher classification performance than multiple-instance learning as well as strong localization results for small lesions through class activation mapping. Entitled “An annotation-free whole-slide training approach to pathological classification of lung cancer types using deep learning,” the article can be accessed online here.
In collaboration with Taipei Medical University Hospital (TMUH), the experiments conducted on a data set of over 9,600 lung cancer WSIs reveal that aetherAI’s proposed method achieves areas under the receiver operating characteristic curve of 0.9594 and 0.9414, respectively, for adenocarcinoma and squamous cell carcinoma classification on the testing set.
“We are excited to see our team’s hard work materialize in a peer-reviewed publication in Nature Communications,” said Joe Yeh, M.D., aetherAI founder and CEO. “The method we propose alleviates the burden of contouring and obtains benefits from scaling up training with numerous WSIs.” Deep learning for digital pathology is hindered by the extremely high spatial resolution of WSIs. Most studies employ patch-based methods, which often require detailed annotation of image patches and typically involve laborious, free-hand contouring on WSIs.
The joint study of aetherAI and TMUH has won the 18th National Innovation Award in 2021, an award created by the Institute for Biotechnology and Medicine Industry (IBMI) in Taiwan which has become the highest honor for biomedical technology corporations and research teams with strong R&D potential.