Search

Purpose This study aims at exploiting artificial intelligence (AI) for the identification, segmentation and quantification of COVID-19 pulmonary lesions. The limited data availability and the annotation quality are relevant factors in training AI-methods. We investigated the effects of using multiple datasets, heterogeneously populated and annotated according to different criteria. Methods We developed an automated analysis pipeline, the LungQuant system, based on a cascade of two U-nets. The first one (U-net$$_1$$ 1 ) is devoted to the identification of the lung parenchyma; the second one (U-net$$_2$$ 2 ) acts on a bounding box enclosing the segmented lungs to identify the areas affected by COVID-19 lesions. Different public datasets were used to train the U-nets and to evaluate their segmentation performances, which have been quantified in terms of the Dice Similarity Coefficients. The accuracy in predicting the CT-Severity Score (CT-SS) of the LungQuant system has been also evaluated. Results Both the volumetric DSC (vDSC) and the accuracy showed a dependency on the annotation quality of the released data samples. On an independent dataset (COVID-19-CT-Seg), both the vDSC and the surface DSC (sDSC) were measured between the masks predicted by LungQuant system and the reference ones. The vDSC (sDSC) values of 0.95±0.01 and 0.66±0.13 (0.95±0.02 and 0.76±0.18, with 5 mm tolerance) were obtained for the segmentation of lungs and COVID-19 lesions, respectively. The system achieved an accuracy of 90% in CT-SS identification on this benchmark dataset. Conclusion We analysed the impact of using data samples with different annotation criteria in training an AI-based quantification system for pulmonary involvement in COVID-19 pneumonia. In terms of vDSC measures, the U-net segmentation strongly depends on the quality of the lesion annotations. Nevertheless, the CT-SS can be accurately predicted on independent test sets, demonstrating the satisfactory generalization ability of the LungQuant.

The aim of this work is to evaluate the potential of combining different computer-aided detection (CADe) methods to increase the actual support for radiologists of automated systems in the identification of pulmonary nodules in CT scans.The outputs of three different CADe systems developed by researchers of the Italian MAGIC-5 collaboration were combined. The systems are: the CAMCADe (based on a Channeler-Ant-Model which segments vessel tree and nodule candidates and a neural classifier), the RGVPCADe (a Region-Growing- Volume-Plateau algorithm detects nodule candidates and a neural network reduces false positives); the VBNACADe (two dedicated procedures, based respectively on a 3D dot-enhancement algorithm and on intersections of pleura surface normals, identifies internal and juxtapleural nodules, and a Voxel-Based-Neural-Approach reduces false positives. A dedicated OsiriX plugin implemented with the Cocoa environments of MacOSX allows annotating nodules and visualizing singles and combined CADe findings.The combined CADe has been tested on thin slice (lower than 2 mm) CTs of the LIDC public research database and the results have been compared with those obtained by the single systems. The FROC (Free Receiver Operating Characteristic) curves show better results than the best of the single approaches.Has been demonstrated that the combination of different approaches offers better results than each single CADe system. A clinical validation of the combined CADe as second reader is being addressed by means of the dedicated OsiriX plugin.