Consistency analysis between deep learning algorithm and radiologists in measuring the volume of lung metastases
Objective To assess the agreement between a deep learning(DL)algorithm and radiologists in measuring lung metastases volume.Methods CT scan images of 57 patients with lung metastases were randomly selected from Xiangtan Central Hospital from June 2019 to June 2023,including 89 solid metastatic nodules.A commercial DL algorithm was used to automatically identify pulmonary metastatic nodules and calculate lung metastases volume by DL(LMV-DL).Simultaneously,two senior radiologists manually outlined the lung metastatic nodules on the lung window and calculated LMV-Radiologist 1 and LMV-Radiologist 2 using the area summation method.The Bland-Altman method was then used to calculate the 95%limits of agreement(95%LoA)between the three groups of LMV,and the agreement was assessed.Results The Bland-Altman method showed that the 95%LoA between the DL algorithm and each radiologist([-758.3 to 416.2]mm³ and[-627.1 to 518.1]mm³,respectively)was wider than the pairwise comparisons between radiologists(-207.1 to 440.2)mm³.Conclusion The DL algorithm demonstrated good consistency with radiologists in measuring LMV and can serve as an automatic measurement method to replace manual measurement,aiding in the clinical management of lung nodules.
Deep learning algorithmRadiologistLung metastasesAssess agreementClinical management