KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Settings

Subject: search.filters.subject.CT scan

Search Results

Now showing 1 - 4 of 4
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    Blood Vessels Detection by Regional-based CNN for CT Scan of Lower Extremities
    (2022-01-01)
    Sakunpaisanwari, Littikrai
    ;
    Yodrabum, Nutcha
    ;
    Sirirapisit, Tanongchai
    ;
    Titijaroonroj, Taravichet
    Blood vessels on computed tomography (CT) scan images are difficult to identify and discriminate between vessels and noise because blood vessels are not only small and shapeless, but its location can also be inconsistent. This is a challenge of object detection. We proposed an automatic blood vessel detection method based on YOLOv3 for object detection from CT scan of lower extremities. This work focused on detecting four main arteries: popliteal, anterior tibial, posterior tibial, and peroneal arteries. To obtain the best architecture for blood vessel detection, we evaluated and compared the performances of seven region-based CNN architectures: Faster R-CNN, Cascade R-CNN, Mask R-CNN, RetinaNet, YOLOv3, CornerNet, and Centernet. Experimental results show that the best architecture was YOLOv3 with precision, recall, and f1-score of 0.982, 0.954, and 0.968, respectively. Good accomplishment of YOLOv3 came from skip connections, multi-scale feature map, and anchor generated by k-means clustering.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    Classification of margin characteristics from 3D pulmonary nodules
    (2017-12-19)
    Paing, May Phu
    ;
    Choomchuay, Somsak
    Detection of pulmonary nodules has played a significant role in lung cancer diagnosis because nodules are the first suspicious symptoms for the likelihood of cancer. Margin characteristics of the pulmonary nodules provide essential radiological features to determine the possibility of malignancy. In general, benign nodules hold quite smooth margins whilst malignant ones hold irregular margins. The main objective of this research is to classify different margin types of pulmonary nodules by observing the 3D structure. Nodule candidates from 2D lung CT slices are segmented firstly and then stacked to form a 3D image. Geometric features of the 3D nodule are extracted and fed into the support vector machine (SVM) classifier to classify the margin types. The proposed method can provide the classification accuracy of 90.9%.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    A computer aided diagnosis system for detection of lung nodules from series of CT slices
    (2017-11-03)
    Paing, May Phu
    ;
    Choomchuay, Somsak
    The proposed system aims to detect the lung nodules from a series of CT scan images. Otsu's thresholding and morphological operations are applied for nodules segmentation. After segmentation, the objects that do not hold the possibility to be nodules are removed. Geometric, histogram as well as texture features are then extracted for benign and malignant nodules classification. Multilayer Perceptron (MLP) is used for classification and the accuracy 95% has been achieved.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    Ground glass opacity (GGO) nodules detection from lung CT scans
    (2017-07-01)
    Paing, May Phu
    ;
    Choomchuay, Somsak
    Ground glass opacity (GGO) nodules have a higher possibility of malignancy compared to other types of nodules appeared in the lung cancer. They are very effortful to detect due to their hazy structures and unclear margins. 65% of lung cancers are missed by the radiologists with the faint appearance of the GGO. Consequently, the detection of GGO is a critical issue and a striving task for the radiologists. This research proposes an automatic detection of GGO remained after the detection of solid opacity. Simple thresholding based on grey levels and mathematical image subtraction are applied for segmentation. Possible false after segmentation are reduced by the support vector machine (SVM). In total 37 GGOs, only 2 are missed by proposed segmentation and the false reduction performance is 94%.