Deep Learning-Based Classification and Semantic Segmentation of Lung Tuberculosis Lesions in Chest X-ray Images.

We present a deep learning (DL) network-based approach for detecting and semantically segmenting two specific types of tuberculosis (TB) lesions in chest X-ray (CXR) images. In the proposed method, we use a basic U-Net model and its enhanced versions to detect, classify, and segment TB lesions in CXR images. The model architectures used in this study are U-Net, Attention U-Net, U-Net++, Attention U-Net++, and pyramid spatial pooling (PSP) Attention U-Net++, which are optimized and compared based on the test results of each model to find the best parameters. Finally, we use four ensemble approaches which combine the top five models to further improve lesion classification and segmentation results. In the training stage, we use data augmentation and preprocessing methods to increase the number and strength of lesion features in CXR images, respectively. Our dataset consists of 110 training, 14 validation, and 98 test images. The experimental results show that the proposed ensemble model achieves a maximum mean intersection-over-union (MIoU) of 0.70, a mean precision rate of 0.88, a mean recall rate of 0.75, a mean F1-score of 0.81, and an accuracy of 1.0, which are all better than those of only using a single-network model. The proposed method can be used by clinicians as a diagnostic tool assisting in the examination of TB lesions in CXR images.

Deep-Learning-Enabled Computer-Aided Diagnosis in the Classification of Pancreatic Cystic Lesions on Confocal Laser Endomicroscopy.

Accurate classification of pancreatic cystic lesions (PCLs) is important to facilitate proper treatment and to improve patient outcomes. We utilized the convolutional neural network (CNN) of VGG19 to develop a computer-aided diagnosis (CAD) system in the classification of subtypes of PCLs in endoscopic ultrasound-guided needle-based confocal laser endomicroscopy (nCLE). From a retrospectively collected 22,424 nCLE video frames (50 videos) as the training/validation set and 11,047 nCLE video frames (18 videos) as the test set, we developed and compared the diagnostic performance of three CNNs with distinct methods of designating the region of interest. The diagnostic accuracy for subtypes of PCLs by CNNs with manual, maximal rectangular, and U-Net algorithm-designated ROIs was 100%, 38.9%, and 66.7% on a per-video basis and 88.99%, 73.94%, and 76.12% on a per-frame basis, respectively. Our per-frame analysis suggested differential levels of diagnostic accuracy among the five subtypes of PCLs, where non-mucinous PCLs (serous cystic neoplasm: 93.11%, cystic neuroendocrine tumor: 84.31%, and pseudocyst: 98%) had higher diagnostic accuracy than mucinous PCLs (intraductal papillary mucinous neoplasm: 84.43% and mucinous cystic neoplasm: 86.1%). Our CNN demonstrated superior specificity compared to the state-of-the-art for the classification of mucinous PCLs (IPMN and MCN), with high specificity (94.3% and 92.8%, respectively) but low sensitivity (46% and 45.2%, respectively). This suggests the complimentary role of CNN-enabled CAD systems, especially for clinically suspected mucinous PCLs.

Miniaturization and image optimization of a full-color holographic display system using a vibrating light guide.

In this study, a miniaturized full-color holographic reconstruction system that uses a single spatial light modulator to achieve full-color image reconstruction was developed. The reconstruction system uses a single light guide for light combination and is therefore less voluminous than conventional reconstruction systems. The experimental results demonstrated that the system had a full-color display, corrected light combination, and eliminated zero-order light. The vibrations of the light guide disrupted the temporal coherence of the laser beam, thus ensuring that the speckle in the reconstructed image was almost imperceptible to the human eye.

Computer-aided diagnosis for identifying and delineating early gastric cancers in magnifying narrow-band imaging.

BACKGROUND AND AIMS: Magnifying narrow-band imaging (M-NBI) is important in the diagnosis of early gastric cancers (EGCs) but requires expertise to master. We developed a computer-aided diagnosis (CADx) system to assist endoscopists in identifying and delineating EGCs. METHODS: We retrospectively collected and randomly selected 66 EGC M-NBI images and 60 non-cancer M-NBI images into a training set and 61 EGC M-NBI images and 20 non-cancer M-NBI images into a test set. After preprocessing and partition, we determined 8 gray-level co-occurrence matrix (GLCM) features for each partitioned 40 × 40 pixel block and calculated a coefficient of variation of 8 GLCM feature vectors. We then trained a support vector machine (SVMLv1) based on variation vectors from the training set and examined in the test set. Furthermore, we collected 2 determined P and Q GLCM feature vectors from cancerous image blocks containing irregular microvessels from the training set, and we trained another SVM (SVMLv2) to delineate cancerous blocks, which were compared with expert-delineated areas for area concordance. RESULTS: The diagnostic performance revealed accuracy of 96.3%, precision (positive predictive value [PPV]) of 98.3%, recall (sensitivity) of 96.7%, and specificity of 95%, at a rate of 0.41 ± 0.01 seconds per image. The performance of area concordance, on a block basis, demonstrated accuracy of 73.8% ± 10.9%, precision (PPV) of 75.3% ± 20.9%, recall (sensitivity) of 65.5% ± 19.9%, and specificity of 80.8% ± 17.1%, at a rate of 0.49 ± 0.04 seconds per image. CONCLUSIONS: This pilot study demonstrates that our CADx system has great potential in real-time diagnosis and delineation of EGCs in M-NBI images.

Computer-aided diagnosis in endoscopy: a novel application toward automatic detection of abnormal lesions on magnifying narrow-band imaging endoscopy in the stomach.

Gastric cancer is the fourth common cancer and the second major cause of cancer death worldwide. Early detection of gastric cancer by endoscopy surveillance is actively investigated to improve patient survival, especially using the newly developed magnifying narrow-band imaging endoscopy in the stomach. However, meticulous examination of the aforementioned images is both time and experience demanding and interpretation could be variable among different doctors, which hindered its widespread application. In this study, we developed a new image analysis system by adopting local binary pattern and vector quantization to perform pattern comparison between known training abnormal images and testing images of magnifying narrow band endoscopy images in the stomach. Our preliminary results demonstrated promising potential for automatically labeled region of interest for endoscopy doctors to focus on abnormal lesions for subsequent targeted biopsy, with the rates of recall 0.46-1.00 and precision 0.39-0.87.