臨床T1/2期乳癌におけるヨウ素マップ由来マルチリージョンラジオミクスおよびマルチモダリティ画像特性を用いた腋窩リンパ節転移の予測

To explore the feasibility of an integrative model incorporating multi-region radiomic features extracted from chest dual-energy CT (DECT)-based iodine maps, clinical parameters, and CT and ultrasonography (US) features of axillary lymph node (ALN), to preoperatively predict ALN metastasis (ALNM) in clinical T1/2 stage breast cancer.

This retrospective study enrolled 197 patients with breast cancer who underwent preoperative contrast-enhanced DECT from March 2021 to May 2022. Radiomic features were extracted from venous-phase iodine maps based on three regions of interests (ROIs): ALN, tumoral and peritumoral regions (2.5 mm around the tumor). Clinical information, CT and US parameters were recorded and evaluated. Eight predictive models were built: 1) A clinical model; 2) CT features-based model; 3) US-based model; 4) tumor-based radiomic model; 5) peritumor-based radiomic model; 6) ALN-based radiomic model; 7) multi-ROIs radiomics model; 8) integrative model. The ALNM prediction performances and clinical usefulness were assessed.

Radiomic signatures derived from ALN, tumor, peritumoral and multi-ROIs achieved AUCs of 0.860, 0.709, 0.747 and 0.890 in the training cohort, and 0.860, 0.676, 0.663, and 0.890 in the testing cohort, respectively. The integrative model incorporating tumor location, T-stage, Ki-67 index, hilus structure, shortest nodal diameter, intranodal vascular pattern, and multi-region radiomic features, demonstrated further increased AUCs of 0.923 and 0.914, with good calibration and clinical benefit.

The model integrating clinical parameters, DECT- and US- reported ALN features, as well as iodine map-derived multi-region radiomic features, could serve as a potential tool to preoperatively predict ALN status.