ADC-Seg: Application of an improved nnU-Net for automatic segmentation of apparent diffusion coefficient images of pulmonary lesions

Motivation: Quantitative analysis of the heterogeneity of PNMs is important in assessing benignity and malignancy and provides a certain basis for the early diagnosis of lung cancer. Goal(s): Automatic segmentation of PNMs by improved nnU-Net model and quantitative analysis to assess benignity and malignancy. Approach: ADC-Seg was constructed by training and testing the improved nnU-Net model with ADC images of 181 patients, and ADC histograms were automatically calculated. Results: The Dice scores of the model were 0.765 and 0.812 for the validation and test sets, and a correlation was observed between the parameters ske, 10th, 15th, and 20th in benign and malignant identification. Impact: The model is capable of accurately segmenting lesions in ADC images and performing quantitative analyses, which will assist imaging physicians in making better diagnoses for assessing the benignity and malignancy of lesions.