Automatic lesion detection and segmentation of 18F-FET PET in gliomas: A full 3D U-Net convolutional neural network study

Abstract

IntroductionAmino-acids positron emission tomography (PET) is increasingly used in the diagnostic workup of patients with gliomas, including differential diagnosis, evaluation of tumor extension, treatment planning and follow-up. Recently, progresses of computer vision and machine learning have been translated for medical imaging. Aim was to demonstrate the feasibility of an automated 18F-fluoro-ethyl-tyrosine (18F-FET) PET lesion detection and segmentation relying on a full 3D U-Net Convolutional Neural Network (CNN).MethodsAll dynamic 18F-FET PET brain image volumes were temporally realigned to the first dynamic acquisition, coregistered and spatially normalized onto the Montreal Neurological Institute template. Ground truth segmentations were obtained using manual delineation and thresholding (1.3 x background). The volumetric CNN was implemented based on a modified Keras implementation of a U-Net library with 3 layers for the encoding and decoding paths. Dice similarity coefficient (DSC) was used as an accuracy measure of segmentation.ResultsThirty-seven patients were included (26 [70%] in the training set and 11 [30%] in the validation set). All 11 lesions were accurately detected with no false positive, resulting in a sensitivity and a specificity for the detection at the tumor level of 100%. After 150 epochs, DSC reached 0.7924 in the training set and 0.7911 in the validation set. After morphological dilatation and fixed thresholding of the predicted U-Net mask a substantial improvement of the DSC to 0.8231 (+ 4.1%) was noted. At the voxel level, this segmentation led to a 0.88 sensitivity [95% CI, 87.1 to, 88.2%] a 0.99 specificity [99.9 to 99.9%], a 0.78 positive predictive value: [76.9 to 78.3%], and a 0.99 negative predictive value [99.9 to 99.9%].ConclusionsWith relatively high performance, it was proposed the first full 3D automated procedure for segmentation of 18F-FET PET brain images of patients with different gliomas using a U-Net CNN architecture.

Artificial intelligence in cancer imaging: Clinical challenges and applications

“…An escalating cascade of studies and methodologies for the semiautomatic and automatic detection of CNS tumors has been published in recent years, largely applied to conventional MR imaging, but also to PET and ultrasound images . Although they are used most frequently in the exploratory and research setting, semiautomatic algorithms have been applied to treatment planning for stereotactic radiosurgery, quantitating the volume of residual tumor after surgery, and tracking tumor growth over time .…”

“…An escalating cascade of studies and methodologies for the semiautomatic and automatic detection of CNS tumors has been published in recent years, largely applied to conventional MR imaging, but also to PET and ultrasound images. [106][107][108][109][110][111][112][113] Although they are used most frequently in the exploratory and research setting, semiautomatic algorithms have been applied to treatment planning for stereotactic radiosurgery, 111 quantitating the volume of residual tumor after surgery, 113 and tracking tumor growth over time. 109 One can envision the benefits of a robust, automatic tumor-detection algorithm in the assessment of patients who have numerous intracranial lesions, such as within the setting of CNS metastases, and their differential growth rate or response to treatment over time.…”

Evaluation of artificial intelligence for detecting periapical pathosis on cone‐beam computed tomography scans

“…This AI method can learn adaptive image characteristics and simultaneously make image classifications (LeCun et al , Shin et al , Tajbakhsh et al , Kim & MacKinnon ). CNNs have been successfully used for automatic assessment of various medical and dental problems, including image‐based automated diagnosis to detect lung and brain lesions (Akkus et al , Song et al , Wang et al , Blanc‐Durand et al ), breast cancer in mammography images (Becker et al ), colorectal polyps and prostate cancer (Wang et al , Byrne et al ), skin cancer (Esteva et al ), diabetic retinopathy in retinal fundus photographs (Gulshan et al ), hip osteoarthritis (Xue et al ) and bone age assessment (Lee et al ). In dentistry, CNNs have been applied to detect carious lesions, periapical lesions, tooth eruption and numbering, vertical root fractures, assess root morphology or periodontal bone loss, dental and jaw pathosis, osteoporosis, and maxillary sinusitis on dental radiographs (Kositbowornchai et al , Miki et al , Ezhov et al , Murata et al , Poedjiastoeti & Suebnukarn , Lee et al ,b, Zakirov et al , Zakirov et al , Chen et al , Ekert et al , Hiraiwa et al , Hwang et al , Krois et al , Tuzoff et al ).…”

Repeatability of Two Artificial Intelligence Approaches for Tumor Segmentation in PET

“…However, only a few studies use AI based segmentation approaches for metabolic active tumor segmentation in PET images. Even more, most studies combine the information of PET and CT images in order to get reliable segmentation results [14] or use some post-processing for an improvement of CNN segmentations [15]. Classi ers classifying each voxel as tumor or non-tumor using textural features of voxel neighborhoods have been used for the segmentation of e.g.…”