Applying Compressed Sensing Volumetric Interpolated Breath-Hold Examination and Spiral Ultrashort Echo Time Sequences for Lung Nodule Detection in MRI

This prospective study aimed to investigate the ability of spiral ultrashort echo time (UTE) and compressed sensing volumetric interpolated breath-hold examination (CS-VIBE) sequences in magnetic resonance imaging (MRI) compared to conventional VIBE and chest computed tomography (CT) in terms of image quality and small nodule detection. Patients with small lung nodules scheduled for video-assisted thoracoscopic surgery (VATS) for lung wedge resection were prospectively enrolled. Each patient underwent non-contrast chest CT and non-contrast MRI on the same day prior to thoracic surgery. The chest CT was performed to obtain a standard reference for nodule size, location, and morphology. The chest MRI included breath-hold conventional VIBE and CS-VIBE with scanning durations of 11 and 13 s, respectively, and free-breathing spiral UTE for 3.5–5 min. The signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and normal structure visualizations were measured to evaluate MRI quality. Nodule detection sensitivity was evaluated on a lobe-by-lobe basis. Inter-reader and inter-modality reliability analyses were performed using the Cohen κ statistic and the nodule size comparison was performed using Bland–Altman plots. Among 96 pulmonary nodules requiring surgery, the average nodule diameter was 7.7 ± 3.9 mm (range: 4–20 mm); of the 73 resected nodules, most were invasive cancer (74%) or pre-invasive carcinoma in situ (15%). Both spiral UTE and CS-VIBE images achieved significantly higher overall image quality scores, SNRs, and CNRs than conventional VIBE. Spiral UTE (81%) and CS-VIBE (83%) achieved a higher lung nodule detection rate than conventional VIBE (53%). Specifically, the nodule detection rate for spiral UTE and CS-VIBE reached 95% and 100% for nodules >8 and >10 mm, respectively. A 90% detection rate was achieved for nodules of all sizes with a part-solid or solid morphology. Spiral UTE and CS-VIBE under-estimated the nodule size by 0.2 ± 1.4 mm with 95% limits of agreement from −2.6 to 2.9 mm and by 0.2 ± 1.7 mm with 95% limits of agreement from −3.3 to 3.5 mm, respectively, compared to the reference CT. In conclusion, chest CT remains the gold standard for lung nodule detection due to its high image resolutions. Both spiral UTE and CS-VIBE MRI could detect small lung nodules requiring surgery and could be considered a potential alternative to chest CT; however, their clinical application requires further investigation.

Designing Chest X-ray Datasets for Improving Lung Nodules Detection Through Convolutional Neural Networks

In this paper, we propose a method for building alternative training datasets for lung nodule detection from plain chest X-ray images. Our aim is to improve the classification quality of a state-of-the-art CNN by just selecting appropriate samples from the existing datasets. The hypothesis of this research is that high quality models need to learn by contrasting very clean images with those containing nodules, specially those difficult to identify by non-expert clinicians. Current chest X-ray datasets mostly include images where more than one pathology exist and/or contain devices like catheters. This is because most samples come from old people which are the usual patients subject to X-ray examinations. In this paper, we evaluate several combinations of samples from existing datasets in the literature. Results show a great gain in performance for some of the evaluated combinations, confirming our hypothesis. The achieved performance of these models allows a considerable speed-up in the screening of patients by radiologist.

Lung Nodule Detection using Segmentation Approach for Computed Tomography Scan Images

Lung disorders have become really common in today’s world due to growing amount of air pollution, our increased exposure to harmful radiations and our unhealthy lifestyles. Hence, the diagnosis of lung disorders has become of paramount importance. The commonly used Thresholding approaches and morphological operations often fail to detect the peripheral pathology bearing areas. Hence, we present the segmentation approach of the lung tissue for computer aided diagnosis system. We use a novel technique for segmentation of lungs from CT scan (Computed Tomography) of the chest or upper torso. The accuracy of analysis and its implication majorly depends on the kind of segmentation technique used. Hence, it is important that the method used is highly reliable and is successful in nodule detection and classification. We use MATLAB and OpenCV libraries to apply segmentation on CT scan images to get the desired output. We have also created a working proprietary user interface called “PULMONIS” for the ease of doctors and patients to upload the CT scan images and get the output after the image processing is done in the backend. Keywords: Lung nodule detection, Image Processing, Computed Tomography, Image Segmentation, Lung Cancer, Contour Segmentation, MATLAB, OpenCV, Computer Vision.