Di Dong

Multi-crop Convolutional Neural Networks for lung nodule malignancy suspiciousness classification

Abstract We investigate the problem of lung nodule malignancy suspiciousness (the likelihood of nodule malignancy) classification using thoracic Computed Tomography (CT) images. Unlike traditional studies primarily relying on cautious nodule segmentation and time-consuming feature extraction, we tackle a more challenging task on directly modeling raw nodule patches and building an end-to-end machine-learning architecture for classifying lung nodule malignancy suspiciousness. We present a Multi-crop Convolutional Neural Network (MC-CNN) to automatically extract nodule salient information by employing a novel multi-crop pooling strategy which crops different regions from convolutional feature maps and then applies max-pooling different times. Extensive experimental results show that the proposed method not only achieves state-of-the-art nodule suspiciousness classification performance, but also effectively characterizes nodule semantic attributes (subtlety and margin) and nodule diameter which are potentially helpful in modeling nodule malignancy.

Real-Time Visualized Freehand 3D Ultrasound Reconstruction Based on GPU

Visualized freehand 3-D ultrasound reconstruction offers to image incremental reconstruction during acquisition and guide users to scan interactively for high-quality volumes. We originally used the graphics processing unit (GPU) to develop a visualized reconstruction algorithm that achieves real-time level. Each newly acquired image was transferred to the memory of the GPU and inserted into the reconstruction volume on the GPU. The partially reconstructed volume was then rendered using GPU-based incremental ray casting. After visualized reconstruction, hole-filling was performed on the GPU to fill remaining empty voxels in the reconstruction volume. We examine the real-time nature of the algorithm using in vitro and in vivo datasets. The algorithm can image incremental reconstruction at speed of 26-58 frames/s and complete 3-D imaging in the acquisition time for the conventional freehand 3-D ultrasound.

Radiomics features of Multiparametric MRI as Novel Prognostic Factors in Advanced Nasopharyngeal Carcinoma

Purpose: To identify MRI-based radiomics as prognostic factors in patients with advanced nasopharyngeal carcinoma (NPC). Experimental Design: One-hundred and eighteen patients (training cohort: n = 88; validation cohort: n = 30) with advanced NPC were enrolled. A total of 970 radiomics features were extracted from T2-weighted (T2-w) and contrast-enhanced T1-weighted (CET1-w) MRI. Least absolute shrinkage and selection operator (LASSO) regression was applied to select features for progression-free survival (PFS) nomograms. Nomogram discrimination and calibration were evaluated. Associations between radiomics features and clinical data were investigated using heatmaps. Results: The radiomics signatures were significantly associated with PFS. A radiomics signature derived from joint CET1-w and T2-w images showed better prognostic performance than signatures derived from CET1-w or T2-w images alone. One radiomics nomogram combined a radiomics signature from joint CET1-w and T2-w images with the TNM staging system. This nomogram showed a significant improvement over the TNM staging system in terms of evaluating PFS in the training cohort (C-index, 0.761 vs. 0.514; P < 2.68 × 10−9). Another radiomics nomogram integrated the radiomics signature with all clinical data, and thereby outperformed a nomogram based on clinical data alone (C-index, 0.776 vs. 0.649; P < 1.60 × 10−7). Calibration curves showed good agreement. Findings were confirmed in the validation cohort. Heatmaps revealed associations between radiomics features and tumor stages. Conclusions: Multiparametric MRI-based radiomics nomograms provided improved prognostic ability in advanced NPC. These results provide an illustrative example of precision medicine and may affect treatment strategies. Clin Cancer Res; 23(15); 4259–69. ©2017 AACR.

Fast Katsevich Algorithm Based on GPU for Helical Cone-Beam Computed Tomography

Katsevich reconstruction algorithm represents a breakthrough for helical cone-beam computed tomography (CT) reconstruction, because it is the first exact cone-beam reconstruction algorithm of filtered backprojection (FBP) type with 1-D shift-invariant filtering. Although FBP-type reconstruction algorithm is effective, 3-D CT reconstruction is time-consuming, and the accelerations of Katsevich algorithm on CPU or cluster have been widely studied. In this paper, Katsevich algorithm is accelerated by using graphics processing unit, including flat-detector and curved-detector geometry in the case of helical orbit. An overscan formula is derived, which helps to avoid unnecessary overscan in practical CT scanning. Based on the overscan formula, a volume-blocking method in device memory is proposed. One advantage of the blocking method is that it can reconstruct large volume with high speed.

Helical optical projection tomography

A new technique termed Helical Optical Projection Tomography (hOPT) has been developed with the aim to overcome some of the limitations of current 3D optical imaging techniques. hOPT is based on Optical Projection Tomography (OPT) with the major difference that there is a translation of the sample in the vertical direction during the image acquisition process, requiring a new approach to image reconstruction. Contrary to OPT, hOPT makes possible to obtain 3D-optical images of intact long samples without imposing limits on the sample length. This has been tested using hOPT to image long murine tissue samples such as spinal cords and large intestines. Moreover, 3D-reconstructed images of the colon of DSS-treated mice, a model for Inflammatory Bowel Disease, allowed the identification of the structural alterations. Finally, the geometry of the hOPT device facilitates the addition of a Selective Plane Illumination Microscopy (SPIM) arm, providing the possibility of delivering high resolution images of selected areas together with complete volumetric information.

Automated Recovery of the Center of Rotation in Optical Projection Tomography in the Presence of Scattering

Finding the center of rotation is an essential step for accurate 3-D reconstruction in optical projection tomography. Unfortunately, current methods are not convenient since they require either prior scanning of a reference phantom, small structures of high intensity existing in the specimen, or active participation during the centering procedure. To solve these problems this paper proposes a fast and automatic center of rotation search method making use of parallel programming in graphics processing units. Our method is based on a two step search approach making use only of those sections of the image with high signal-to-noise ratio. We have tested this method both in nonscattering ex vivo samples and in in vivo specimens with a considerable contribution of scattering such as Drosophila melanogaster pupae, recovering in all cases the center of rotation with a precision 1/4 pixel or less.

The development and validation of a CT-based radiomics signature for the preoperative discrimination of stage I-II and stage III-IV colorectal cancer

Objectives To investigative the predictive ability of radiomics signature for preoperative staging (I-IIvs.III-IV) of primary colorectal cancer (CRC). Methods This study consisted of 494 consecutive patients (training dataset: n=286; validation cohort, n=208) with stage I–IV CRC. A radiomics signature was generated using LASSO logistic regression model. Association between radiomics signature and CRC staging was explored. The classification performance of the radiomics signature was explored with respect to the receiver operating characteristics(ROC) curve. Results The 16-feature-based radiomics signature was an independent predictor for staging of CRC, which could successfully categorize CRC into stage I-II and III-IV (p <0.0001) in training and validation dataset. The median of radiomics signature of stage III-IV was higher than stage I-II in the training and validation dataset. As for the classification performance of the radiomics signature in CRC staging, the AUC was 0.792(95%CI:0.741-0.853) with sensitivity of 0.629 and specificity of 0.874. The signature in the validation dataset obtained an AUC of 0.708(95%CI:0.698-0.718) with sensitivity of 0.611 and specificity of 0.680. Conclusions A radiomics signature was developed and validated to be a significant predictor for discrimination of stage I-II from III-IV CRC, which may serve as a complementary tool for the preoperative tumor staging in CRC.

Early detection of liver cancer based on bioluminescence tomography

As a new modality of molecular imaging, bioluminescence imaging has been widely used in tumor detection and drug evaluation. However, BLI cannot present the depth of information for internal diseases such as a liver tumor in situ or a lung tumor in situ. In this paper, we describe a bioluminescence tomography (BLT) method based on the bioluminescent intensity attenuation calibration and applied it to the early detection of liver cancer in situ. In comparison with BLT without calibration, this method could improve the reconstruction accuracy by more than 10%. In comparison with micro-computed tomography and other traditional imaging modalities, this method can detect a liver tumor at a very early stage and provide reliable location information.

Learning from Experts: Developing Transferable Deep Features for Patient-Level Lung Cancer Prediction

Due to recent progress in Convolutional Neural Networks (CNNs), developing image-based CNN models for predictive diagnosis is gaining enormous interest. However, to date, insufficient imaging samples with truly pathological-proven labels impede the evaluation of CNN models at scale. In this paper, we formulate a domain-adaptation framework that learns transferable deep features for patient-level lung cancer malignancy prediction. The presented work learns CNN-based features from a large discovery set (2272 lung nodules) with malignancy likelihood labels involving multiple radiologists’ assessments, and then tests the transferable predictability of these CNN-based features on a diagnosis-definite set (115 cases) with true pathologically-proven lung cancer labels. We evaluate our approach on the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset, where both human expert labeling information on cancer malignancy likelihood and a set of pathologically-proven malignancy labels were provided. Experimental results demonstrate the superior predictive performance of the transferable deep features on predicting true patient-level lung cancer malignancy (Acc = 70.69 %, AUC = 0.66), which outperforms a nodule-level CNN model (Acc = 65.38 %, AUC = 0.63) and is even comparable to that of using the radiologists’ knowledge (Acc = 72.41 %, AUC = 0.76). The proposed model can largely reduce the demand for pathologically-proven data, holding promise to empower cancer diagnosis by leveraging multi-source CT imaging datasets.

In-vivo Optical Tomography of Small Scattering Specimens: time-lapse 3D imaging of the head eversion process in Drosophila melanogaster

Even though in vivo imaging approaches have witnessed several new and important developments, specimens that exhibit high light scattering properties such as Drosophila melanogaster pupae are still not easily accessible with current optical imaging techniques, obtaining images only from subsurface features. This means that in order to obtain 3D volumetric information these specimens need to be studied either after fixation and a chemical clearing process, through an imaging window - thus perturbing physiological development -, or during early stages of development when the scattering contribution is negligible. In this paper we showcase how Optical Projection Tomography may be used to obtain volumetric images of the head eversion process in vivo in Drosophila melanogaster pupae, both in control and headless mutant specimens. Additionally, we demonstrate the use of Helical Optical Projection Tomography (hOPT) as a tool for high throughput 4D-imaging of several specimens simultaneously.