Fig. 5: Analysis of tumor boundary via AI-ECM and demonstration of translational potential via in vivo imaging of mouse model harboring human lung cancer.

a ROC curve of one vs. all SVM model. b Vocabulary distributions for collagen and elastin fibers in boundary. c Representative H&E stained histological image of lung cancer tissues. Scale bar: 1 mm. Solid black line marks the border between cancerous (upper left) and normal (lower right) regions, and dashed box represents one MPM image field. d The MPM image corresponding to the dashed box in (c). Scale bar: 100 μm. CRI maps of collagen (e) and elastin (f) fibers corresponding to the MPM image. g The merged collagen and elastin CRI map. h Establishment of murine orthotopic model through implantation of human lung cancer cells. i Schematic of in vivo imaging protocol. j Typical 3D MPM stack obtained from in vivo imaging. Scale bar: 100 μm. k Boxplots showing comparison of fiber metrics between normal and cancer samples, including collagen directional variance (***p = 1.0e−4), collagen waviness (***p = 2.0e−4), collagen local coverage (***p = 7.4e−7), SC (***p = 1.0e−6), elastin directional variance (***p = 1.4e−4), elastin waviness (***p = 2.0e−4), elastin local coverage (***p = 5.0e−6), and elastin thickness (***p = 2.2e−5). A total of 12 MPM volumes for normal tissues and 12 ones for cancer tissues are captured with n = 4 mice for each group. DV directional variance, Wav waviness, LC local coverage, Thi thickness, Ela elastin, Col collagen. Two-tailed Mann–Whitney tests were used to determine significant differences.