Artificial intelligence and radiomics enhance the positive predictive value of digital chest tomosynthesis for lung cancer detection within SOS clinical trial

Abstract

Objectives: within this investigation we investigated several approaches to enhance the positive predictive value (PPV) of chest digital tomosynthesis (DTS) in the lung cancer detection Methods: the investigation was carried out within the SOS clinical trial (NCT03645018) for lung cancer screening with DTS. Lung nodules were identified by visual analysis and then classified using the diameter and the radiological aspect of the nodule following a modified lung-RADS classification.Haralick texture features were extracted from the segmented nodules. Both semantic variables and radiomics features were used to build a predictive model using two approaches: logistic regression model on a sub-set of variables selected with backward feature selection or machine learning using the whole sub-set of variables. We used two machine learning methods: a Random Forest and a neural network. Machine learning methods were applied to a training set and validated on a test set. Methods were compared using diagnostic accuracy metrics.Results: binary visual analysis had a good sensitivity (0.95) but a low PPV (0.14). Lung-RADS classification increased the PPV (0.19) but with an unacceptable low sensitivity (0.65). Analogously, logistic regression showed a mildly increased PPV (0.22) and a low sensitivity (0.67). Random Forest demonstrated a low accuracy with a moderate PPV (0.40) but with a dramatically low sensitivity (0.30). Neural network demonstrated to be the best predictor with a nearly perfect PPV (0.95) and a high sensitivity (0.90).Conclusions: among the various technique to reduce the false positive rates of DTS the neural network demonstrated a very high PPV. The use of visual analysis along with neural network could help radiologists to depict a follow-up strategy after a positive DTS.

Machine Learning-Based Differentiation of Nontuberculous Mycobacteria Lung Disease and Pulmonary Tuberculosis Using CT Images

“…Interestingly, both features have been reported in related clinical studies. For instance, the minor axis length of shape is important in the detection of clinically significant prostate cancer in multiparametric MR images [ 34 ], and the zone entropy of GLSZM reflects the areas with different gray intensities within the nodules for lung cancer detection [ 35 ]. However, it should be noted that both features cannot be perceived directly, and thus, accurate segmentation of the bronchiectasis regions becomes indispensable.…”

"Multimodal" Imaging Performance and Clinical Application of Coronavirus Disease 2019 (COVID-19)

“…In recent years, many studies had also corroborated the important role of DTS in the discovery and follow-up of chest nodular lesions, and even the discovery and evaluation of lung cancer [16]. With the development and expansion of DTS technology, its role in the neck, breast [17], the thoracic spine, abdomen and skeletal system becomes more and more signi cant [18].…”

Artificial intelligence applications in medical imaging: A review of the medical physics research in Italy

“…Random forest (RF), a popular concept in ML, are based on a large set of randomly generated decision trees which are trained individually. After training, the prediction is made for all the individual trees and the most frequently selected class is taken as a final result [26][27][28][29][30].…”

“…ML showed a good accuracy in distinguishing benign from malignant nodules in low-dose CT screening, and outperformed the clinical standard in an independent validation cohort [10]. A lung cancer detection model using RF and ANN for radiomic features selection was developed for chest digital tomosynthesis, [27], and the least absolute shrinkage and selection operator (LASSO) logistic regression modelwas used to predict positive lymph nodes [11]. DL was also applied with promising results to multimodality PET/CT lung images for the classification of lung cancer lesions as T1-T2 or T3-T4 [60].…”