Salta al contenuto principale
Passa alla visualizzazione normale.

ROBERTO LAGALLA

Focal breast lesion characterization according to the BI-RADS US lexicon: role of a computer-aided decision-making support

  • Autori: Bartolotta, Tommaso Vincenzo; Orlando, Alessia; Cantisani, Vito; Matranga, Domenica; Ienzi, Raffele; Cirino, Alessandra; Amato, Francesco; Di Vittorio, Maria Laura; Midiri, Massimo; Lagalla, Roberto
  • Anno di pubblicazione: 2018
  • Tipologia: Articolo in rivista (Articolo in rivista)
  • OA Link: http://hdl.handle.net/10447/325885

Abstract

Objectives: to assess the diagnostic performance of a computer-guided decision- making software (S-Detect) in the US characterization of focal breast lesions (FBLs), according to the radiologist's experience. Materials and Methods: 300 FBLs (size: 2.6-47.2 mm; mean: 13.2 mm) in 255 patients (mean age: 51 years) were prospectively assessed in consensus according to BIRADS US lexicon by two experienced radiologists without and with S-Detect; to evaluate intra and inter-observer agreement, the same 300 FBLs were independently evaluated by two residents at baseline and after 3 months. Results: 120/300 (40%) FBLs were malignant, 2/300 (0.7%) high-risk and 178/300 (59.3%) benign. Experts review showed a not significant increase in Sensitivity, Specificity, PPV and NPV with S-Detect (97.5%, 86.5%, 83.2%, 98.1%) than without (91.8%, 81.5%, 77.2%, 93.6%) (p>0.05), as confirmed by ROC curve analysis (0.95 with and 0.92 without S-Detect [p=0.0735]). A significant higher area under the ROC curve (0.88) with S-Detect than without (0.85) was found for Resident #1 (p=0.0067) and Resident #2 (0.83 without and 0.87 with S-Detect [p=0.0302]). Intra-observer agreement (k score) improved with S-Detect from 0.69 to 0.78 for Resident #1 (p>0.05) and from 0.69 to 0.81 for Resident #2 (p>0.05). Inter-observer agreement improved with S-Detect from 0.67 to 0.7 (baseline; p>0.05) and from 0.63 to 0.77 (after 3 months; p>0.05). According to S-Detect-guided re-classification, 27/64 (42.2%) FBLs underwent a correct change in clinical management, 25/64 (39.1%) FBLs underwent no change and 12/68 (18.7%) FBLs underwent an uncorrect change. Conclusion: S-Detect can be used as an effective tool for classification of FBLs, especially for less experienced physicians.