Researchers led by presenter Dr. Alessandro Furlan of the Pittsburgh Liver Center in Pittsburgh, PA, retrospectively gathered a dataset of 819 patients with liver fibrosis and who had received portal-venous phase CT exams for liver fibrosis staging between January 2015 and January 2022. Of these, 623 were used to train algorithms and 196 were set aside for testing.
The researchers developed three AI models. The first was based only on liver analysis, while the second only analyzed the spleen. The last algorithm, which combined both liver and spleen analysis, yielded the best results.
In testing on 196 patients, the combined model outperformed both radiologists and serum tests. Furthermore, the diagnostic performance of the combined model wasn’t influenced by patient characteristics, pathology, and CT data.
“Application of this segmentation and classification algorithm may help clinicians in preoperative planning for liver surgery, therapeutic effect evaluation, and [predicting] the prognosis of chronic liver disease, without excess burden,” the authors wrote.
Want to know more? Attend this talk on Sunday afternoon.
