Radiomics to increase the effectiveness of lung cancer screening programs. Radiolung preliminary results.
2022; Elsevier BV; Linguagem: Inglês
10.1183/13993003.congress-2022.1498
ISSN1872-8332
AutoresAntoni Rosell Gratacós, Sonia Baeza, Samuel García, José Luís Mate, Ignasi Guasch, Isabel Nogueira, Ignasi García-Olivé, Gabriel Saboia de Araújo Torres, C Sànchez-Ramos, D Gil,
Tópico(s)Lung Cancer Diagnosis and Treatment
ResumoIntroduction: Main lung cancer screening clinical trials demonstrated a 20-25% mortality reduction but rendered 19.7-27.3% false-positive nodules in the chest CT. Radiomics can provide discriminatory capacity beyond what is perceived by the naked eye. Goal: Establish a radiomic signature of pulmonary nodules (PN) to distinguish malignancy from benignity. Patients: Prospective observational study with PNs studied and resected according to usual clinical practice. Method: CT images are sent to the Computer Vision Center, and the nodules segmented. Gray Level Co-occurrence Matrix radiomic based texture features that significatively correlate with malignancy are extracted. These variables are used to train a neural network with an architecture optimized according to the diagnosis of the nodule. The model has been trained with 8 benign PNs and 43 malignant PNs from our hospital, and subsequently validated with 6 benign and 21 malignant PNs from our hospital and from a public database. Results: 51 PNs of 22.68mm (range 3-45mm), 32 men, mean age 69 years, 41 smokers or ex-smokers were analyzed. The pathological results were: 32 (62.7%) ADK, 7 (13.7%) SCC, 3 (5.8%) NSCLC, 1 (1.9%) carcinoid; 2 (3.9%) nonspecific necrosis, 6 (11.7%) benign tumors. The diagnostic accuracy of our hybrid system is 96.30%, 100% sensitivity, specificity of 83.3%. Conclusions: In our sample, the application of a hybrid radiomic system achieves high diagnostic accuracy (96.3%) to detect malignant nodules on chest CT. External validation in a lung cancer screening program is needed. Funded by : ACMCiB, BRN, Fundació Ramon Pla, Lung Ambition Alliance
Referência(s)