"AI could also differentiate between leiomyosarcoma and liposarcoma in 84% of sarcomas tested, while radiologists were unable to tell the difference in 35% of c

"AI could also differentiate between leiomyosarcoma and liposarcoma in 84% of sarcomas tested, while radiologists were unable to tell the difference in 35% of cases." - What is the ground truth based on? - Are the biopsy results coming from the same 89 cases? - What is the signal AI based system is picking up that humans have failed to from different modalities? - Why is the biopsy even considered as a good test when it exhibits such poor performance?