Members of our team have contributed to an important multicenter study, recently published, titled "Inter-vendor Performance of Deep Learning in Segmenting Acute Ischemic Lesions on Diffusion-Weighted Imaging." This research addresses the challenge of applying deep learning (DL) techniques across different MRI scanner brands to segment acute ischemic lesions accurately.

The study analyzed DWI data from patients with acute ischemic lesions collected from six centers, utilizing scanners from Siemens and GE. Two distinct datasets, A (n = 2,986) and B (n = 3,951), underwent model training, validation, and testing. The developed neural networks, Models A and B, were initially trained on their respective datasets and later fine-tuned using transfer learning techniques across the datasets.

The outcomes demonstrated promising results, with median Dice scores of 0.858 and 0.857 for internal tests on Models A and B, matching the performance of experienced radiologists and showcasing no inferiority. However, in external tests, initial models showed slightly lower performance compared to a radiologist, which was then mitigated through fine-tuning, resulting in improved scores of 0.832 and 0.846, respectively. These results highlight the potential of deep learning models to achieve cross-vendor generalizability and clinical applicability through transfer learning.

At Hevi AI, we are committed to continuing our scientific endeavors, enhancing the accuracy and usability of AI technologies in medical imaging. For a more detailed exploration of this study, please visit [Nature](https://www.nature.com/articles/s41598-021-91467-x).