Bone metastasis classification using whole body images from prostate cancer patients based on convolutional neural networks application

SPECT/CT, PET/CT, and PET/MRI for Response Assessment of Bone Metastases

Recent developments in hybrid SPECT/CT systems and the use of cadmium-zinc-telluride (CZT) detectors have improved the diagnostic accuracy of bone scintigraphy. These advancements have paved the way for novel quantitative approaches to accurate and reproducible treatment monitoring of bone metastases. PET/CT imaging using [18F]F-FDG and [18F]F-NaF have shown promising clinical utility in bone metastases assessment and monitoring response to therapy and prediction of treatment response in a broad range of malignancies. Additionally, specific tumor-targeting tracers like [99mTc]Tc-PSMA, [68Ga]Ga-PSMA, or [11C]C- or [18F]F-Choline revealed high diagnostic performance for early assessment and prognostication of bone metastases, particularly in prostate cancer. PET/MRI appears highly accurate imaging modality, but has associated limitations notably, limited availability, more complex logistics and high installation costs. Advances in artificial intelligence (Al) seem to improve the accuracy of imaging modalities and provide an assistant role in the evaluation of treatment response of bone metastases.

AI-assisted Analysis to Facilitate Detection of Humeral Lesions on Chest Radiographs

Purpose To develop an artificial intelligence (AI) system for humeral tumor detection on chest radiographs (CRs) and evaluate the impact on reader performance. Materials and Methods In this retrospective study, 14 709 CRs (January 2000 to December 2021) were collected from 13 468 patients, including CT-proven normal (n = 13 116) and humeral tumor (n = 1593) cases. The data were divided into training and test groups. A novel training method called false-positive activation area reduction (FPAR) was introduced to enhance the diagnostic performance by focusing on the humeral region. The AI program and 10 radiologists were assessed using holdout test set 1, wherein the radiologists were tested twice (with and without AI test results). The performance of the AI system was evaluated using holdout test set 2, comprising 10 497 normal images. Receiver operating characteristic analyses were conducted for evaluating model performance. Results FPAR application in the AI program improved its performance compared with a conventional model based on the area under the receiver operating characteristic curve (0.87 vs 0.82, P = .04). The proposed AI system also demonstrated improved tumor localization accuracy (80% vs 57%, P < .001). In holdout test set 2, the proposed AI system exhibited a false-positive rate of 2%. AI assistance improved the radiologists' sensitivity, specificity, and accuracy by 8.9%, 1.2%, and 3.5%, respectively (P < .05 for all). Conclusion The proposed AI tool incorporating FPAR improved humeral tumor detection on CRs and reduced false-positive results in tumor visualization. It may serve as a supportive diagnostic tool to alert radiologists about humeral abnormalities. Keywords: Artificial Intelligence, Conventional Radiography, Humerus, Machine Learning, Shoulder, Tumor Supplemental material is available for this article. © RSNA, 2024.

Stability of standardized uptake values for quantitative bone SPECT for jawbone lesions: a single-center cross-sectional study

BACKGROUND

The long time required for bone uptake of radiopharmaceutical material after injection for bone scintigraphy is a burden for patients with poor health. Thus, to assess whether the uptake time could be reduced for single-photon emission computed tomography (SPECT) of the jawbone, this study evaluated differences in maximum standardized uptake values (SUVmax) within patients using SPECT imaging at 2 and 3 hours after radiopharmaceutical injection.

METHODS

A total of 33 patients undergoing treatment or in post-treatment follow-up for medication-related osteonecrosis of the jaw, who visited our hospital between July 2020 and August 2021 and could receive SPECT twice on the same day, were enrolled in the study. Patients were injected with technetium-99 m hydroxymethylene diphosphonate (Tc-99 m HMDP) intravenously. The SUVmax for healthy parietal bones and jawbone lesions were calculated from the SPECT images using quantitative analysis software, and the SUVmax were compared between 2- and 3-hour uptake times.

RESULTS

After exclusion, 30 patients were included in the study. In the 2-hour and 3-hour images, the median SUVmax of the parietal bones were 1.90 and 1.81, respectively, and those of the jawbone lesions were 9.25 and 9.39, respectively. The limits of agreement (LOA) ranged from - 0.33 to 0.25 in the parietal bones, and the %LOA ranged from - 9.8 to 17.3% in the jawbone lesions, showing high equivalence between the two uptake durations. The SUVmax showed no clinical differences between the 2- and 3-hour uptake durations for Tc-99 m HMDP SPECT of the jawbone.

CONCLUSIONS

The results of this study justify a 2-3-hour uptake window when performing quantitative SPECT of the jawbone. Therefore, the minimum uptake time can potentially be reduced to only 2 hours.

Bone Metastasis in Prostate Cancer: Bone Scan Versus PET Imaging

Prostate cancer is the second most common cause of malignancy among men, with bone metastasis being a significant source of morbidity and mortality in advanced cases. Detecting and treating bone metastasis at an early stage is crucial to improve the quality of life and survival of prostate cancer patients. This objective strongly relies on imaging studies. While CT and MRI have their specific utilities, they also possess certain drawbacks. Bone scintigraphy, although cost-effective and widely available, presents high false-positive rates. The emergence of PET/CT and PET/MRI, with their ability to overcome the limitations of standard imaging methods, offers promising alternatives for the detection of bone metastasis. Various radiotracers targeting cell division activity or cancer-specific membrane proteins, as well as bone seeking agents, have been developed and tested. The use of positron-emitting isotopes such as fluorine-18 and gallium-68 for labeling allows for a reduced radiation dose and unaffected biological properties. Furthermore, the integration of artificial intelligence (AI) and radiomics techniques in medical imaging has shown significant advancements in reducing interobserver variability, improving accuracy, and saving time. This article provides an overview of the advantages and limitations of bone scan using SPECT and SPECT/CT and PET imaging methods with different radiopharmaceuticals and highlights recent developments in hybrid scanners, AI, and radiomics for the identification of prostate cancer bone metastasis using molecular imaging.

Computer-aided diagnosis of skeletal metastases in multi-parametric whole-body MRI

The confident detection of metastatic bone disease is essential to improve patients' comfort and increase life expectancy. Multi-parametric magnetic resonance imaging (MRI) has been successfully used for monitoring of metastatic bone disease, allowing for comprehensive and holistic evaluation of the total tumour volume and treatment response assessment. The major challenges of radiological reading of whole-body MRI come from the amount of data to be reviewed and the scattered distribution of metastases, often of complex shapes. This makes bone lesion detection and quantification demanding for a radiologist and prone to error. Additionally, whole-body MRI are often corrupted with multiple spatial and intensity distortions, which further degrade the performance of image reading and image processing algorithms. In this work we propose a fully automated computer-aided diagnosis system for the detection and segmentation of metastatic bone disease using whole-body multi-parametric MRI. The system consists of an extensive image preprocessing pipeline aiming at enhancing the image quality, followed by a deep learning framework for detection and segmentation of metastatic bone disease. The system outperformed state-of-the-art methodologies, achieving a detection sensitivity of 63% with a mean of 6.44 false positives per image, and an average lesion Dice coefficient of 0.53. A provided ablation study performed to investigate the relative importance of image preprocessing shows that introduction of region of interest mask and spatial registration have a significant impact on detection and segmentation performance in whole-body MRI. The proposed computer-aided diagnosis system allows for automatic quantification of disease infiltration and could provide a valuable tool during radiological examination of whole-body MRI.

Performance of deep learning models for response evaluation on whole-body bone scans in prostate cancer

OBJECTIVE

We aimed to develop deep learning classifiers for assessing therapeutic response on bone scans of patients with prostate cancer.

METHODS

A set of 3791 consecutive bone scans coupled with their last previous scan (1528 patients) was evaluated. Bone scans were labeled as "progression" or "nonprogression" on the basis of clinical reports and image review. A 2D-convolutional neural network architecture was trained with three different preprocessing methods: 1) no preprocessing (Raw), 2) spatial normalization (SN), and 3) spatial and count normalization (SCN). Data were allocated into training, validation, and test sets in the ratio of 72:8:20, with the 20% independent test set rotating all scans over a five-fold testing procedure. A Grad-CAM algorithm was employed to generate class activation maps to visualize the lesions contributing to the decision. Diagnostic performance was compared using area under the receiver operating characteristics curves (AUCs).

RESULTS

The data consisted of 791 scans labeled as "progression" and 3000 scans labeled as "nonprogression." The AUCs of the classifiers were 0.632-0.710 on the Raw dataset, were significantly higher with the use of SN at 0.784-0.854 (p < 0.001 for Raw versus SN), and higher still with SCN at 0.954-0.979 (p < 0.001 for SN versus SCN). Class activation maps of the SCN model visualized lesions contributing to the model's decision of progression.

CONCLUSION

With preprocessing of spatial and count normalization, our deep learning model achieved excellent performance in classifying the therapeutic response of bone scans in patients with prostate cancer.

Imaging signs and the qualitative diagnosis of solitary rib lesions using 99mtechnetium-methylene diphosphonate whole-body bone imaging in patients with a malignant tumor

Background

The aim of this study was to summarize the valuable information for qualitative diagnosis by investigating the imaging signs from the whole-body bone imaging of solitary rib lesions.

Methods

A retrospective analysis was conducted of the data from 313 patients with malignant tumors and solitary rib lesions identified using whole-body bone imaging in Department of Nuclear Medicine of Central South University Xiangya School Affiliated Haikou Hospital between January 2015 and December 2017. Based on the final comprehensive diagnosis of the rib lesions, the patients were divided into a bone metastasis group, fracture group, other benign lesions group, and an uncertain group, and the characteristic imaging changes in rib lesions in each group were explored.

Results

(I) Significant differences were identified among the 4 groups (P<0.001) in the distribution of lesions in the anterior, posterior, and lateral ribs and proximal costal cartilage. The fracture group had the highest proportion of lesions in the anterior ribs (99/121, 81.8%) and proximal costal cartilage (74.4%, 90/121). (II) Significant differences were detected in morphology, concentration, boundaries, and radioactivity distribution among the 4 groups of patients (P<0.001). The bone metastasis group had the highest proportion of lesions appearing as stripes (35/67, 52.2%), and the fracture group had the highest proportion of lesions appearing as spots (94.2%, 114/121) and the lowest proportion appearing as stripes (3/121, 2.5%). (III) Significant differences were found in the longitudinal diameter, transverse diameter, aspect ratio, and tumor-to-normal tissue ratio between the 4 groups (P<0.001). The longitudinal diameter (27.8±16.0 mm) and aspect ratio (1.9±1.0) of the bone metastasis group were the highest, whereas the longitudinal diameter (15.2±3.9 mm) and aspect ratio (1.0±0.2) of the fracture group were the smallest.

Conclusions

This study revealed that different types of solitary rib lesions had relatively characteristic imaging signs in whole-body bone imaging.

Artificial intelligence-based analysis of whole-body bone scintigraphy: The quest for the optimal deep learning algorithm and comparison with human observer performance

PURPOSE

Whole-body bone scintigraphy (WBS) is one of the most widely used modalities in diagnosing malignant bone diseases during the early stages. However, the procedure is time-consuming and requires vigour and experience. Moreover, interpretation of WBS scans in the early stages of the disorders might be challenging because the patterns often reflect normal appearance that is prone to subjective interpretation. To simplify the gruelling, subjective, and prone-to-error task of interpreting WBS scans, we developed deep learning (DL) models to automate two major analyses, namely (i) classification of scans into normal and abnormal and (ii) discrimination between malignant and non-neoplastic bone diseases, and compared their performance with human observers.

MATERIALS AND METHODS

After applying our exclusion criteria on 7188 patients from three different centers, 3772 and 2248 patients were enrolled for the first and second analyses, respectively. Data were split into two parts, including training and testing, while a fraction of training data were considered for validation. Ten different CNN models were applied to single- and dual-view input (posterior and anterior views) modes to find the optimal model for each analysis. In addition, three different methods, including squeeze-and-excitation (SE), spatial pyramid pooling (SPP), and attention-augmented (AA), were used to aggregate the features for dual-view input models. Model performance was reported through area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, and specificity and was compared with the DeLong test applied to ROC curves. The test dataset was evaluated by three nuclear medicine physicians (NMPs) with different levels of experience to compare the performance of AI and human observers.

RESULTS

DenseNet121_AA (DensNet121, with dual-view input aggregated by AA) and InceptionResNetV2_SPP achieved the highest performance (AUC = 0.72) for the first and second analyses, respectively. Moreover, on average, in the first analysis, Inception V3 and InceptionResNetV2 CNN models and dual-view input with AA aggregating method had superior performance. In addition, in the second analysis, DenseNet121 and InceptionResNetV2 as CNN methods and dual-view input with AA aggregating method achieved the best results. Conversely, the performance of AI models was significantly higher than human observers for the first analysis, whereas their performance was comparable in the second analysis, although the AI model assessed the scans in a drastically lower time.

CONCLUSION

Using the models designed in this study, a positive step can be taken toward improving and optimizing WBS interpretation. By training DL models with larger and more diverse cohorts, AI could potentially be used to assist physicians in the assessment of WBS images.

Systematic Review of Tumor Segmentation Strategies for Bone Metastases

Purpose: To investigate the segmentation approaches for bone metastases in differentiating benign from malignant bone lesions and characterizing malignant bone lesions. Method: The literature search was conducted in Scopus, PubMed, IEEE and MedLine, and Web of Science electronic databases following the guidelines of Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). A total of 77 original articles, 24 review articles, and 1 comparison paper published between January 2010 and March 2022 were included in the review. Results: The results showed that most studies used neural network-based approaches (58.44%) and CT-based imaging (50.65%) out of 77 original articles. However, the review highlights the lack of a gold standard for tumor boundaries and the need for manual correction of the segmentation output, which largely explains the absence of clinical translation studies. Moreover, only 19 studies (24.67%) specifically mentioned the feasibility of their proposed methods for use in clinical practice. Conclusion: Development of tumor segmentation techniques that combine anatomical information and metabolic activities is encouraging despite not having an optimal tumor segmentation method for all applications or can compensate for all the difficulties built into data limitations.

A comprehensive survey of deep learning research on medical image analysis with focus on transfer learning

This survey aims to identify commonly used methods, datasets, future trends, knowledge gaps, constraints, and limitations in the field to provide an overview of current solutions used in medical image analysis in parallel with the rapid developments in transfer learning (TL). Unlike previous studies, this survey grouped the last five years of current studies for the period between January 2017 and February 2021 according to different anatomical regions and detailed the modality, medical task, TL method, source data, target data, and public or private datasets used in medical imaging. Also, it provides readers with detailed information on technical challenges, opportunities, and future research trends. In this way, an overview of recent developments is provided to help researchers to select the most effective and efficient methods and access widely used and publicly available medical datasets, research gaps, and limitations of the available literature.

Classification and diagnostic prediction of breast cancer metastasis on clinical data using machine learning algorithms

Metastatic Breast Cancer (MBC) is one of the primary causes of cancer-related deaths in women. Despite several limitations, histopathological information about the malignancy is used for the classification of cancer. The objective of our study is to develop a non-invasive breast cancer classification system for the diagnosis of cancer metastases. The anaconda-Jupyter notebook is used to develop various python programming modules for text mining, data processing, and Machine Learning (ML) methods. Utilizing classification model cross-validation criteria, including accuracy, AUC, and ROC, the prediction performance of the ML models is assessed. Welch Unpaired t-test was used to ascertain the statistical significance of the datasets. Text mining framework from the Electronic Medical Records (EMR) made it easier to separate the blood profile data and identify MBC patients. Monocytes revealed a noticeable mean difference between MBC patients as compared to healthy individuals. The accuracy of ML models was dramatically improved by removing outliers from the blood profile data. A Decision Tree (DT) classifier displayed an accuracy of 83% with an AUC of 0.87. Next, we deployed DT classifiers using Flask to create a web application for robust diagnosis of MBC patients. Taken together, we conclude that ML models based on blood profile data may assist physicians in selecting intensive-care MBC patients to enhance the overall survival outcome.

BS-80K: The first large open-access dataset of bone scan images

BACKGROUND

Radionuclide bone scanning is one of the most common tools in the inspection of bone metastasis. Conventionally, the analysis of bone scan image is derived from manual diagnosing. However, this task requires extensive subjective diagnostic experience and is extremely time-consuming. To this end, a series of studies concerning computer-aided diagnosis via machine learning tools have been proposed. Although some inspiring progress has been achieved, the implemented bone scan image datasets in these research areas are generally too small, private or non-general, which limits their practical significance and impedes the follow-up research.

METHOD

To address this issue, we present a large, publicly available and general dataset consisting of 82544 bone scan images associated with 3247 patients from West China Hospital, named BS-80K. In BS-80K, each patient provides two whole bone scan images corresponding to the anterior view (ANT) and the posterior view (POST). For each view, there are 13 region-wise slices of the body parts susceptible to bone metastasis. Based on an authorized original labeling criterion, labels annotated by experienced specialists are offered with the images. Moreover, within each whole body image, multiple bounding boxes containing suspectable hot spots and their annotations are supplied as well. All images in BS-80K have been de-identified to protect patients' privacy.

RESULTS

Based on 6 popular deep learning models for classification and object detection, we provide the benchmark for a number of computer-aided medical tasks, including general bone metastasis prediction and object detection for whole body images, and specific bone metastasis prediction for different body parts. According to extensive experiments, the adopted classification models achieve remarkable results in accuracy and specificity (around 95%) on most metastasis prediction tasks, which are approximate to the average ability of corresponding specialists. As for the object detection task, the best average precision of the adopted models reaches 0.2484 and the lowest is 0.1334.

DISCUSSION

Through the comparison of metastasis prediction performance between the benchmark and related work, we observe that the widely used models trained by BS-80K achieve significantly better results than the elaborately designed models trained by smaller datasets. This indicates that with the large amount of data, BS-80K has great potential to galvanize the research about computer-aided analysis on bone scan image.

CONCLUSION

To the best of our knowledge, BS-80K is the first large publicly available dataset of bone scanning, which favors a wide range of research on computer-aided bone metastasis diagnosis. The full dataset is now available at https://drive.google.com/drive/folders/1DOBkLXgQeREQjF-nQIGNBBzPCb5s7RNu?usp=sharing.

Semi-supervised segmentation of metastasis lesions in bone scan images

To develop a deep image segmentation model that automatically identifies and delineates lesions of skeletal metastasis in bone scan images, facilitating clinical diagnosis of lung cancer–caused bone metastasis by nuclear medicine physicians. A semi-supervised segmentation model is proposed, comprising the feature extraction subtask and pixel classification subtask. During the feature extraction stage, cascaded layers which include the dilated residual convolution, inception connection, and feature aggregation learn the hierarchal representations of low-resolution bone scan images. During the pixel classification stage, each pixel is first classified into categories in a semi-supervised manner, and the boundary of pixels belonging to an individual lesion is then delineated using a closed curve. Experimental evaluation conducted on 2,280 augmented samples (112 original images) demonstrates that the proposed model performs well for automated segmentation of metastasis lesions, with a score of 0.692 for DSC if the model is trained using 37% of the labeled samples. The self-defined semi-supervised segmentation model can be utilized as an automated clinical tool to detect and delineate metastasis lesions in bone scan images, using only a few manually labeled image samples. Nuclear medicine physicians need only attend to those segmented lesions while ignoring the background when they diagnose bone metastasis using low-resolution images. More images of patients from multiple centers are typically needed to further improve the scalability and performance of the model via mitigating the impacts of variability in size, shape, and intensity of bone metastasis lesions.

Emerging Applications of Deep Learning in Bone Tumors: Current Advances and Challenges

Deep learning is a subfield of state-of-the-art artificial intelligence (AI) technology, and multiple deep learning-based AI models have been applied to musculoskeletal diseases. Deep learning has shown the capability to assist clinical diagnosis and prognosis prediction in a spectrum of musculoskeletal disorders, including fracture detection, cartilage and spinal lesions identification, and osteoarthritis severity assessment. Meanwhile, deep learning has also been extensively explored in diverse tumors such as prostate, breast, and lung cancers. Recently, the application of deep learning emerges in bone tumors. A growing number of deep learning models have demonstrated good performance in detection, segmentation, classification, volume calculation, grading, and assessment of tumor necrosis rate in primary and metastatic bone tumors based on both radiological (such as X-ray, CT, MRI, SPECT) and pathological images, implicating a potential for diagnosis assistance and prognosis prediction of deep learning in bone tumors. In this review, we first summarized the workflows of deep learning methods in medical images and the current applications of deep learning-based AI for diagnosis and prognosis prediction in bone tumors. Moreover, the current challenges in the implementation of the deep learning method and future perspectives in this field were extensively discussed.

Radiomics and artificial intelligence in prostate cancer: new tools for molecular hybrid imaging and theragnostics

In prostate cancer (PCa), the use of new radiopharmaceuticals has improved the accuracy of diagnosis and staging, refined surveillance strategies, and introduced specific and personalized radioreceptor therapies. Nuclear medicine, therefore, holds great promise for improving the quality of life of PCa patients, through managing and processing a vast amount of molecular imaging data and beyond, using a multi-omics approach and improving patients’ risk-stratification for tailored medicine. Artificial intelligence (AI) and radiomics may allow clinicians to improve the overall efficiency and accuracy of using these “big data” in both the diagnostic and theragnostic field: from technical aspects (such as semi-automatization of tumor segmentation, image reconstruction, and interpretation) to clinical outcomes, improving a deeper understanding of the molecular environment of PCa, refining personalized treatment strategies, and increasing the ability to predict the outcome. This systematic review aims to describe the current literature on AI and radiomics applied to molecular imaging of prostate cancer.

Convolutional neural networks for detection of transthyretin amyloidosis in 2D scintigraphy images

BACKGROUND

Transthyretin amyloidosis (ATTR) is a progressive disease which can be diagnosed non-invasively using bone avid [^99mTc]-labeled radiotracers. Thus, ATTR is also an occasional incidental finding on bone scintigraphy. In this study, we trained convolutional neural networks (CNN) to automatically detect and classify ATTR from scintigraphy images. The study population consisted of 1334 patients who underwent [^99mTc]-labeled hydroxymethylene diphosphonate (HMDP) scintigraphy and were visually graded using Perugini grades (grades 0-3). A total of 47 patients had visual grade ≥ 2 which was considered positive for ATTR. Two custom-made CNN architectures were trained to discriminate between the four Perugini grades of cardiac uptake. The classification performance was compared to four state-of-the-art CNN models.

RESULTS

Our CNN models performed better than, or equally well as, the state-of-the-art models in detection and classification of cardiac uptake. Both models achieved area under the curve (AUC) ≥ 0.85 in the four-class Perugini grade classification. Accuracy was good in detection of negative vs. positive ATTR patients (grade < 2 vs grade ≥ 2, AUC > 0.88) and high-grade cardiac uptake vs. other patients (grade < 3 vs. grade 3, AUC = 0.94). Maximum activation maps demonstrated that the automated deep learning models were focused on detecting the myocardium and not extracardiac features.

CONCLUSION

Automated convolutional neural networks can accurately detect and classify different grades of cardiac uptake on bone scintigraphy. The CNN models are focused on clinically relevant image features. Automated screening of bone scintigraphy images using CNN could improve the early diagnosis of ATTR.

Risk factors associated with skeletal-related events following discontinuation of denosumab treatment among patients with bone metastases from solid tumors: A real-world machine learning approach

•

This study investigated SRE risk factors after densomuab treatment discontinuation.

•

An unbiased machine learning approach was developed to evaluate >60 variables.

•

Prior SREs and short denosumab treatment duration were primary risk factors.

•

The results can guide denosumab persistence decisions and improve patient outcomes.

Background

Clinical practice guidelines recommend the use of bone-targeting agents for preventing skeletal-related events (SREs) among patients with bone metastases from solid tumors. The anti-RANKL monoclonal antibody denosumab is approved for the prevention of SREs in patients with bone metastases from solid tumors. However, real-world data are lacking on the impact of individual risk factors for SREs, specifically in the context of denosumab discontinuation.

Purpose

We aim to identify risk factors associated with SRE incidence following denosumab discontinuation using a machine learning approach to help profile patients at a higher risk of developing SREs following discontinuation of denosumab treatment.

Methods

Using the Optum PanTher Electronic Health Record repository, patients diagnosed with incident bone metastases from primary solid tumors between January 1, 2007, and September 1, 2019, were evaluated for inclusion in the study. Eligible patients received ≥ 2 consecutive 120 mg denosumab doses on a 4-week (± 14 days) schedule with a minimum follow-up of ≥ 1 year after the last denosumab dose, or an SRE occurring between days 84 and 365 after denosumab discontinuation. Extreme gradient boosting was used to develop an SRE risk prediction model evaluated on a test dataset. Multiple variables associated with patient demographics, comorbidities, laboratory values, treatments, and denosumab exposures were examined as potential factors for SRE risk using Shapley Additive Explanations (SHAP). Univariate analyses on risk factors with the highest importance from pooled and tumor-specific models were also conducted.

Results

A total of 1,414 adult cancer patients (breast: 40%, prostate: 30%, lung: 13%, other: 17%) were eligible, of whom 1,133 (80%) were assigned to model training and 281 (20%) to model evaluation. The median age at inclusion was 67 (range, 19–89) years with a median duration of denosumab treatment of 253 (range, 88–2,726) days; 490 (35%) patients experienced ≥ 1 SRE 83 days after denosumab discontinuation. Meaningful model performance was evaluated by an area under the receiver operating curve score of 77% and an F1 score of 62%; model precision was 60%, with 63% sensitivity and 78% specificity. SHAP identified several significant factors for the tumor-agnostic and tumor-specific models that predicted an increased SRE risk following denosumab discontinuation, including prior SREs, shorter denosumab treatment duration, ≥ 4 clinic visits per month with at least one hospitalization (all-cause) event from the baseline period up to discontinuation of denosumab, younger age at bone metastasis, shorter time to denosumab initiation from bone metastasis, and prostate cancer.

Conclusion

This analysis showed a higher cumulative number of SREs, prior SREs relative to denosumab initiation, a higher number of hospital visits, and a shorter denosumab treatment duration as significant factors that are associated with an increased SRE risk after discontinuation of denosumab, in both the tumor-agnostic and tumor-specific models. Our machine learning approach to SRE risk factor identification reinforces treatment guidance on the persistent use of denosumab and has the potential to help clinicians better assess a patient’s need to continue denosumab treatment and improve patient outcomes.

Automated detection of lung cancer-caused metastasis by classifying scintigraphic images using convolutional neural network with residual connection and hybrid attention mechanism

Background

Whole-body bone scan is the widely used tool for surveying bone metastases caused by various primary solid tumors including lung cancer. Scintigraphic images are characterized by low specificity, bringing a significant challenge to manual analysis of images by nuclear medicine physicians. Convolutional neural network can be used to develop automated classification of images by automatically extracting hierarchal features and classifying high-level features into classes.

Results

Using convolutional neural network, a multi-class classification model has been developed to detect skeletal metastasis caused by lung cancer using clinical whole-body scintigraphic images. The proposed method consisted of image aggregation, hierarchal feature extraction, and high-level feature classification. Experimental evaluations on a set of clinical scintigraphic images have shown that the proposed multi-class classification network is workable for automated detection of lung cancer-caused metastasis, with achieving average scores of 0.7782, 0.7799, 0.7823, 0.7764, and 0.8364 for accuracy, precision, recall, F-1 score, and AUC value, respectively.

Conclusions

The proposed multi-class classification model can not only predict whether an image contains lung cancer-caused metastasis, but also differentiate between subclasses of lung cancer (i.e., adenocarcinoma and non-adenocarcinoma). On the context of two-class (i.e., the metastatic and non-metastatic) classification, the proposed model obtained a higher score of 0.8310 for accuracy metric.

Artificial intelligence applied to musculoskeletal oncology: a systematic review

Developments in artificial intelligence have the potential to improve the care of patients with musculoskeletal tumors. We performed a systematic review of the published scientific literature to identify the current state of the art of artificial intelligence applied to musculoskeletal oncology, including both primary and metastatic tumors, and across the radiology, nuclear medicine, pathology, clinical research, and molecular biology literature. Through this search, we identified 252 primary research articles, of which 58 used deep learning and 194 used other machine learning techniques. Articles involving deep learning have mostly involved bone scintigraphy, histopathology, and radiologic imaging. Articles involving other machine learning techniques have mostly involved transcriptomic analyses, radiomics, and clinical outcome prediction models using medical records. These articles predominantly present proof-of-concept work, other than the automated bone scan index for bone metastasis quantification, which has translated to clinical workflows in some regions. We systematically review and discuss this literature, highlight opportunities for multidisciplinary collaboration, and identify potentially clinically useful topics with a relative paucity of research attention. Musculoskeletal oncology is an inherently multidisciplinary field, and future research will need to integrate and synthesize noisy siloed data from across clinical, imaging, and molecular datasets. Building the data infrastructure for collaboration will help to accelerate progress towards making artificial intelligence truly useful in musculoskeletal oncology.

BID-Net: An Automated System for Bone Invasion Detection Occurring at Stage T4 in Oral Squamous Carcinoma Using Deep Learning

Detection of the presence and absence of bone invasion by the tumor in oral squamous cell carcinoma (OSCC) patients is very significant for their treatment planning and surgical resection. For bone invasion detection, CT scan imaging is the preferred choice of radiologists because of its high sensitivity and specificity. In the present work, deep learning algorithm based model, BID-Net, has been proposed for the automation of bone invasion detection. BID-Net performs the binary classification of CT scan images as the images with bone invasion and images without bone invasion. The proposed BID-Net model has achieved an outstanding accuracy of 93.62%. The model is also compared with six Transfer Learning models like VGG16, VGG19, ResNet-50, MobileNetV2, DenseNet-121, ResNet-101 and BID-Net outperformed over the other models. As there exists no previous studies on bone invasion detection using Deep Learning models, so the results of the proposed model have been validated from the experts of practitioner radiologists, S.M.S. hospital, Jaipur, India.

Automated detection of skeletal metastasis of lung cancer with bone scans using convolutional nuclear network

Bone scan is widely used for surveying bone metastases caused by various solid tumors. Scintigraphic images are characterized by inferior spatial resolution, bringing a significant challenge to manual analysis of images by nuclear medicine physicians. We present in this work a new framework for automatically classifying scintigraphic images collected from patients clinically diagnosed with lung cancer. The framework consists of data preparation and image classification. In the data preparation stage, data augmentation is used to enlarge the dataset, followed by image fusion and thoracic region extraction. In the image classification stage, we use a self-defined convolutional neural network consisting of feature extraction, feature aggregation, and feature classification sub-networks. The developed multi-class classification network can not only predict whether a bone scan image contains bone metastasis but also tell which subcategory of lung cancer that a bone metastasis metastasized from is present in the image. Experimental evaluations on a set of clinical bone scan images have shown that the proposed multi-class classification network is workable for automated classification of metastatic images, with achieving average scores of 0.7392, 0.7592, 0.7242, and 0.7292 for accuracy, precision, recall, and F-1 score, respectively.

Does COVID-19 Clinical Status Associate with Outcome Severity? An Unsupervised Machine Learning Approach for Knowledge Extraction

Since the beginning of the COVID-19 pandemic, 195 million people have been infected and 4.2 million have died from the disease or its side effects. Physicians, healthcare scientists and medical staff continuously try to deal with overloaded hospital admissions, while in parallel, they try to identify meaningful correlations between the severity of infected patients with their symptoms, comorbidities and biomarkers. Artificial intelligence (AI) and machine learning (ML) have been used recently in many areas related to COVID-19 healthcare. The main goal is to manage effectively the wide variety of issues related to COVID-19 and its consequences. The existing applications of ML to COVID-19 healthcare are based on supervised classifications which require a labeled training dataset, serving as reference point for learning, as well as predefined classes. However, the existing knowledge about COVID-19 and its consequences is still not solid and the points of common agreement among different scientific communities are still unclear. Therefore, this study aimed to follow an unsupervised clustering approach, where prior knowledge is not required (tabula rasa). More specifically, 268 hospitalized patients at the First Propaedeutic Department of Internal Medicine of AHEPA University Hospital of Thessaloniki were assessed in terms of 40 clinical variables (numerical and categorical), leading to a high-dimensionality dataset. Dimensionality reduction was performed by applying a principal component analysis (PCA) on the numerical part of the dataset and a multiple correspondence analysis (MCA) on the categorical part of the dataset. Then, the Bayesian information criterion (BIC) was applied to Gaussian mixture models (GMM) in order to identify the optimal number of clusters under which the best grouping of patients occurs. The proposed methodology identified four clusters of patients with similar clinical characteristics. The analysis revealed a cluster of asymptomatic patients that resulted in death at a rate of 23.8%. This striking result forces us to reconsider the relationship between the severity of COVID-19 clinical symptoms and the patient's mortality.

Detection of Bone Metastases on Bone Scans through Image Classification with Contrastive Learning

Patients with bone metastases have poor prognoses. A bone scan is a commonly applied diagnostic tool for this condition. However, its accuracy is limited by the nonspecific character of radiopharmaceutical accumulation, which indicates all-cause bone remodeling. The current study evaluated deep learning techniques to improve the efficacy of bone metastasis detection on bone scans, retrospectively examining 19,041 patients aged 22 to 92 years who underwent bone scans between May 2011 and December 2019. We developed several functional imaging binary classification deep learning algorithms suitable for bone scans. The presence or absence of bone metastases as a reference standard was determined through a review of image reports by nuclear medicine physicians. Classification was conducted with convolutional neural network-based (CNN-based), residual neural network (ResNet), and densely connected convolutional networks (DenseNet) models, with and without contrastive learning. Each set of bone scans contained anterior and posterior images with resolutions of 1024 × 256 pixels. A total of 37,427 image sets were analyzed. The overall performance of all models improved with contrastive learning. The accuracy, precision, recall, F1 score, area under the receiver operating characteristic curve, and negative predictive value (NPV) for the optimal model were 0.961, 0.878, 0.599, 0.712, 0.92 and 0.965, respectively. In particular, the high NPV may help physicians safely exclude bone metastases, decreasing physician workload, and improving patient care.

Radiomics for Identification and Prediction in Metastatic Prostate Cancer: A Review of Studies

Metastatic Prostate Cancer (mPCa) is associated with a poor patient prognosis. mPCa spreads throughout the body, often to bones, with spatial and temporal variations that make the clinical management of the disease difficult. The evolution of the disease leads to spatial heterogeneity that is extremely difficult to characterise with solid biopsies. Imaging provides the opportunity to quantify disease spread. Advanced image analytics methods, including radiomics, offer the opportunity to characterise heterogeneity beyond what can be achieved with simple assessment. Radiomics analysis has the potential to yield useful quantitative imaging biomarkers that can improve the early detection of mPCa, predict disease progression, assess response, and potentially inform the choice of treatment procedures. Traditional radiomics analysis involves modelling with hand-crafted features designed using significant domain knowledge. On the other hand, artificial intelligence techniques such as deep learning can facilitate end-to-end automated feature extraction and model generation with minimal human intervention. Radiomics models have the potential to become vital pieces in the oncology workflow, however, the current limitations of the field, such as limited reproducibility, are impeding their translation into clinical practice. This review provides an overview of the radiomics methodology, detailing critical aspects affecting the reproducibility of features, and providing examples of how artificial intelligence techniques can be incorporated into the workflow. The current landscape of publications utilising radiomics methods in the assessment and treatment of mPCa are surveyed and reviewed. Associated studies have incorporated information from multiple imaging modalities, including bone scintigraphy, CT, PET with varying tracers, multiparametric MRI together with clinical covariates, spanning the prediction of progression through to overall survival in varying cohorts. The methodological quality of each study is quantified using the radiomics quality score. Multiple deficits were identified, with the lack of prospective design and external validation highlighted as major impediments to clinical translation. These results inform some recommendations for future directions of the field.

Progress in diagnosis of bone metastasis of prostate cancer

The diagnosis of bone metastasis of prostate cancer (PC) is of great significance to the treatment and prognosis of patients with PC.Bone scan is the most commonly used in the early diagnosis of bone metastasis, but its specificity is low and there is a high false positive.In recent years, with the in-depth study of the application of CT, MRI, emission computed tomography (ECT), positron emission computed tomography/computed tomography (PET/CT) and deep learning algorithm-convolutional neural networks (CNN) in the diagnosis of bone metastasis, the combined application of various auxiliary parameters in the diagnosis of bone metastasis has significantly been improved. The therapeutic effect of PC patients with bone metastasis can also be evaluated, which is expected to achieve the treatment of bone metastasis as well as diagnosis. By systematically expounding the research progress of the above-mentioned techniques in the diagnosis of bone metastasis, it can provide clinicians with new methods for the diagnosis of bone metastasis and improve the diagnostic efficiency for bone metastasis.

Extraction of metastasis hotspots in a whole-body bone scintigram based on bilateral asymmetry

PURPOSE

A hotspot of bone metastatic lesion in a whole-body bone scintigram is often observed as left-right asymmetry. The purpose of this study is to present a network to evaluate bilateral difference of a whole-body bone scintigram, and to subsequently integrate it with our previous network that extracts the hotspot from a pair of anterior and posterior images.

METHODS

Input of the proposed network is a pair of scintigrams that are the original one and the flipped version with respect to body axis. The paired scintigrams are processed by a butterfly-type network (BtrflyNet). Subsequently, the output of the network is combined with the output of another BtrflyNet for a pair of anterior and posterior scintigrams by employing a convolutional layer optimized using training images.

RESULTS

We evaluated the performance of the combined networks, which comprised two BtrflyNets followed by a convolutional layer for integration, in terms of accuracy of hotspot extraction using 1330 bone scintigrams of 665 patients with prostate cancer. A threefold cross-validation experiment showed that the number of false positive regions was reduced from 4.30 to 2.13 for anterior and 4.71 to 2.62 for posterior scintigrams on average compared with our previous model.

CONCLUSIONS

This study presented a network for hotspot extraction of bone metastatic lesion that evaluates bilateral difference of a whole-body bone scintigram. When combining the network with the previous network that extracts the hotspot from a pair of anterior and posterior scintigrams, the false positives were reduced by nearly half compared to our previous model.

Multiclass classification of whole-body scintigraphic images using a self-defined convolutional neural network with attention modules

Purpose

A self‐defined convolutional neural network is developed to automatically classify whole‐body scintigraphic images of concern (i.e., the normal, metastasis, arthritis, and thyroid carcinoma), automatically detecting diseases with whole‐body bone scintigraphy.

Methods

A set of parameter transformation operations are first used to augment the original dataset of whole‐body bone scintigraphic images. A hybrid attention mechanism including the spatial and channel attention module is then introduced to develop a deep classification network, Dscint, which consists of eight weight layers, one hybrid attention module, two normalization modules, two fully connected layers, and one softmax layer.

Results

Experimental evaluations conducted on a set of whole‐body scintigraphic images show that the proposed deep classification network, Dscint, performs well for automated detection of diseases by classifying the images of concerns, achieving the accuracy, precision, recall, specificity, and F‐1 score of 0.9801, 0.9795, 0.9791, 0.9933, and 0.9792, respectively, on the test data in the augmented dataset. A comparative analysis of Dscint and several classical deep classification networks (i.e., AlexNet, ResNet, VGGNet, DenseNet, and Inception‐v4) reveals that our self‐defined network, Dscint, performs best on classifying whole‐body scintigraphic images on the same dataset.

Conclusions

The self‐defined deep classification network, Dscint, can be utilized to automatically determine whether a whole‐body scintigraphic image is either normal or contains diseases of concern. Specifically, better performance of Dscint is obtained on images with lesions that are present in relatively fixed locations like thyroid carcinoma than those with lesions occurring in nonfixed locations of bone tissue.

An Iterative Algorithm for Semisupervised Classification of Hotspots on Bone Scintigraphies of Patients with Prostate Cancer

Prostate cancer (PCa) is the second most diagnosed cancer in men. Patients with PCa often develop metastases, with more than 80% of this metastases occurring in bone. The most common imaging technique used for screening, diagnosis and follow-up of disease evolution is bone scintigraphy, due to its high sensitivity and widespread availability at nuclear medicine facilities. To date, the assessment of bone scans relies solely on the interpretation of an expert physician who visually assesses the scan. Besides this being a time consuming task, it is also subjective, as there is no absolute criteria neither to identify bone metastases neither to quantify them by a straightforward and universally accepted procedure. In this paper, a new algorithm for the false positives reduction of automatically detected hotspots in bone scintigraphy images is proposed. The motivation relies in the difficulty of building a fully annotated database. In this way, our algorithm is a semisupervised method that works in an iterative way. The ultimate goal is to provide the physician with a fast, precise and reliable tool to quantify bone scans and evaluate disease progression and response to treatment. The algorithm is tested in a set of bone scans manually labeled according to the patient's medical record. The achieved classification sensitivity, specificity and false negative rate were 63%, 58% and 37%, respectively. Comparison with other state-of-the-art classification algorithms shows superiority of the proposed method.

dSPIC: a deep SPECT image classification network for automated multi-disease, multi-lesion diagnosis

BACKGROUND

Functional imaging especially the SPECT bone scintigraphy has been accepted as the effective clinical tool for diagnosis, treatment, evaluation, and prevention of various diseases including metastasis. However, SPECT imaging is brightly characterized by poor resolution, low signal-to-noise ratio, as well as the high sensitivity and low specificity because of the visually similar characteristics of lesions between diseases on imaging findings.

METHODS

Focusing on the automated diagnosis of diseases with whole-body SPECT scintigraphic images, in this work, a self-defined convolutional neural network is developed to survey the presence or absence of diseases of concern. The data preprocessing mainly including data augmentation is first conducted to cope with the problem of limited samples of SPECT images by applying the geometric transformation operations and generative adversarial network techniques on the original SPECT imaging data. An end-to-end deep SPECT image classification network named dSPIC is developed to extract the optimal features from images and then to classify these images into classes, including metastasis, arthritis, and normal, where there may be multiple diseases existing in a single image.

RESULTS

A group of real-world data of whole-body SPECT images is used to evaluate the self-defined network, obtaining a best (worst) value of 0.7747 (0.6910), 0.7883 (0.7407), 0.7863 (0.6956), 0.8820 (0.8273) and 0.7860 (0.7230) for accuracy, precision, sensitivity, specificity, and F-1 score, respectively, on the testing samples from the original and augmented datasets.

CONCLUSIONS

The prominent classification performance in contrast to other related deep classifiers including the classical AlexNet network demonstrates that the built deep network dSPIC is workable and promising for the multi-disease, multi-lesion classification task of whole-body SPECT bone scintigraphy images.

Diagnostic performance of deep learning models for detecting bone metastasis on whole-body bone scan in prostate cancer

PURPOSE

We evaluated the performance of deep learning classifiers for bone scans of prostate cancer patients.

METHODS

A total of 9113 consecutive bone scans (5342 prostate cancer patients) were initially evaluated. Bone scans were labeled as positive/negative for bone metastasis using clinical reports and image review for ground truth diagnosis. Two different 2D convolutional neural network (CNN) architectures were proposed: (1) whole body-based (WB) and (2) tandem architectures integrating whole body and local patches, here named as "global-local unified emphasis" (GLUE). Both models were trained using abundant (72%:8%:20% for training:validation:test sets) and limited training data (10%:40%:50%). The allocation of test sets was rotated across all images: therefore, fivefold and twofold cross-validation test results were available for abundant and limited settings, respectively.

RESULTS

A total of 2991 positive and 6142 negative bone scans were used as input. For the abundant training setting, the receiver operating characteristics curves of both the GLUE and WB models indicated excellent diagnostic ability in terms of the area under the curve (GLUE: 0.936-0.955, WB: 0.933-0.957, P > 0.05 in four of the fivefold tests). The overall accuracies of the GLUE and WB models were 0.900 and 0.889, respectively. With the limited training setting, the GLUE models showed significantly higher AUCs than the WB models (0.894-0.908 vs. 0.870-0.877, P < 0.0001).

CONCLUSION

Our 2D-CNN models accurately classified bone scans of prostate cancer patients. While both showed excellent performance with the abundant dataset, the GLUE model showed higher performance than the WB model in the limited data setting.

Predicting Peritoneal Metastasis of Gastric Cancer Patients Based on Machine Learning

OBJECTIVE

The aim is to explore the prediction effect of 5 machine learning algorithms on peritoneal metastasis of gastric cancer.

METHODS

1080 patients with postoperative gastric cancer were divided into a training group and test group according to the ratio of 7:3. The model of peritoneal metastasis was established by using 5 machine learning (gbm(Light Gradient Boosting Machine), GradientBoosting, forest, Logistic and DecisionTree). Python pair was used to analyze the machine learning algorithm. Gbm algorithm is used to show the weight proportion of each variable to the result.

RESULT

Correlation analysis showed that tumor size and depth of invasion were positively correlated with the recurrence of patients after gastric cancer surgery. The results of the gbm algorithm showed that the top 5 important factors were albumin, platelet count, depth of infiltration, preoperative hemoglobin and weight, respectively. In training group: Among the 5 algorithm models, the accuracy of GradientBoosting and gbm was the highest (0.909); the AUC values of the 5 algorithms are gbm (0.938), GradientBoosting (0.861), forest (0.796), Logistic(0.741) and DecisionTree(0.712) from high to low. In the test group: among the 5 algorithm models, the accuracy of forest, DecisionTree and gbm was the highest (0.907); AUC values ranged from high to low to gbm (0.745), GradientBoosting (0.725), forest (0.696), Logistic (0.680) and DecisionTree (0.657).

CONCLUSION

Machine learning can predict the peritoneal metastasis in patients with gastric cancer.

Automatic Diagnosis of Coronary Artery Disease in SPECT Myocardial Perfusion Imaging Employing Deep Learning

Focusing on coronary artery disease (CAD) patients, this research paper addresses the problem of automatic diagnosis of ischemia or infarction using single-photon emission computed tomography (SPECT) (Siemens Symbia S Series) myocardial perfusion imaging (MPI) scans and investigates the capabilities of deep learning and convolutional neural networks. Considering the wide applicability of deep learning in medical image classification, a robust CNN model whose architecture was previously determined in nuclear image analysis is introduced to recognize myocardial perfusion images by extracting the insightful features of an image and use them to classify it correctly. In addition, a deep learning classification approach using transfer learning is implemented to classify cardiovascular images as normal or abnormal (ischemia or infarction) from SPECT MPI scans. The present work is differentiated from other studies in nuclear cardiology as it utilizes SPECT MPI images. To address the two-class classification problem of CAD diagnosis, achieving adequate accuracy, simple, fast and efficient CNN architectures were built based on a CNN exploration process. They were then employed to identify the category of CAD diagnosis, presenting its generalization capabilities. The results revealed that the applied methods are sufficiently accurate and able to differentiate the infarction or ischemia from healthy patients (overall classification accuracy = 93.47% ± 2.81%, AUC score = 0.936). To strengthen the findings of this study, the proposed deep learning approaches were compared with other popular state-of-the-art CNN architectures for the specific dataset. The prediction results show the efficacy of new deep learning architecture applied for CAD diagnosis using SPECT MPI scans over the existing ones in nuclear medicine.

Artificial intelligence in molecular imaging

AI has, to varying degrees, affected all aspects of molecular imaging, from image acquisition to diagnosis. During the last decade, the advent of deep learning in particular has transformed medical image analysis. Although the majority of recent advances have resulted from neural-network models applied to image segmentation, a broad range of techniques has shown promise for image reconstruction, image synthesis, differential-diagnosis generation, and treatment guidance. Applications of AI for drug design indicate the way forward for using AI to facilitate molecular-probe design, which is still in its early stages. Deep-learning models have demonstrated increased efficiency and image quality for PET reconstruction from sinogram data. Generative adversarial networks (GANs), which are paired neural networks that are jointly trained to generate and classify images, have found applications in modality transformation, artifact reduction, and synthetic-PET-image generation. Some AI applications, based either partly or completely on neural-network approaches, have demonstrated superior differential-diagnosis generation relative to radiologists. However, AI models have a history of brittleness, and physicians and patients may not trust AI applications that cannot explain their reasoning. To date, the majority of molecular-imaging applications of AI have been confined to research projects, and are only beginning to find their ways into routine clinical workflows via commercialization and, in some cases, integration into scanner hardware. Evaluation of actual clinical products will yield more realistic assessments of AI’s utility in molecular imaging.

Deep Learning Based Staging of Bone Lesions From Computed Tomography Scans

In this study, we formulated an efficient deep learning-based classification strategy for characterizing metastatic bone lesions using computed tomography scans (CTs) of prostate cancer patients. For this purpose, 2,880 annotated bone lesions from CT scans of 114 patients diagnosed with prostate cancer were used for training, validation, and final evaluation. These annotations were in the form of lesion full segmentation, lesion type and labels of either benign or malignant. In this work, we present our approach in developing the state-of-the-art model to classify bone lesions as benign or malignant, where (1) we introduce a valuable dataset to address a clinically important problem, (2) we increase the reliability of our model by patient-level stratification of our dataset following lesion-aware distribution at each of the training, validation, and test splits, (3) we explore the impact of lesion texture, morphology, size, location, and volumetric information on the classification performance, (4) we investigate the functionality of lesion classification using different algorithms including lesion-based average 2D ResNet-50, lesion-based average 2D ResNeXt-50, 3D ResNet-18, 3D ResNet-50, as well as the ensemble of 2D ResNet-50 and 3D ResNet-18. For this purpose, we employed a train/validation/test split equal to 75%/12%/13% with several data augmentation methods applied to the training dataset to avoid overfitting and to increase reliability. We achieved an accuracy of 92.2% for correct classification of benign vs. malignant bone lesions in the test set using an ensemble of lesion-based average 2D ResNet-50 and 3D ResNet-18 with texture, volumetric information, and morphology having the greatest discriminative power respectively. To the best of our knowledge, this is the highest ever achieved lesion-level accuracy having a very comprehensive data set for such a clinically important problem. This level of classification performance in the early stages of metastasis development bodes well for clinical translation of this strategy.

Lesion-Based Bone Metastasis Detection in Chest Bone Scintigraphy Images of Prostate Cancer Patients Using Pre-Train, Negative Mining, and Deep Learning

This study aimed to explore efficient ways to diagnose bone metastasis early using bone scintigraphy images through negative mining, pre-training, the convolutional neural network, and deep learning. We studied 205 prostate cancer patients and 371 breast cancer patients and used bone scintigraphy data from breast cancer patients to pre-train a YOLO v4 with a false-positive reduction strategy. With the pre-trained model, transferred learning was applied to prostate cancer patients to build a model to detect and identify metastasis locations using bone scintigraphy. Ten-fold cross validation was conducted. The mean sensitivity and precision rates for bone metastasis location detection and classification (lesion-based) in the chests of prostate patients were 0.72 ± 0.04 and 0.90 ± 0.04, respectively. The mean sensitivity and specificity rates for bone metastasis classification (patient-based) in the chests of prostate patients were 0.94 ± 0.09 and 0.92 ± 0.09, respectively. The developed system has the potential to provide pre-diagnostic reports to aid in physicians' final decisions.

Comparison of irregular flux viewer system with BONENAVI version for identification of Tc-99m MDP whole body bone scan metastasis images

The Tc-99m methylene diphosphonate (MDP) whole body bone scan (WBBS) has been widely accepted as a method of choice for the initial diagnosis of bone and joint changes in patients with oncologic diseases. The WBBS has shown high sensitivity but relatively low specificity due to bone variation. This study aims to use the self-developing irregular flux viewer (IFV) system to predict possible bone lesions in planar WBBS. The study uses gradient vector flow (GVF) and self-organizing map (SOM) methods to analyze the blood fluid-dynamics and evaluate hot points. The evaluation includes a selection of 368 patients with bone metastasis from prostate cancer, lung cancer and breast cancer. Finally, we compare IFV values with BONENAVI version data. BONENAVI is a computer-assisted diagnosis system that analyzes bone scintigraphy automatically. The analysis shows that the IFV system achieves sensitivities of 93% for prostate cancer, 91% for breast cancer, and 83% for lung cancer, respectively. On the other hand, our proposed approach achieves a higher sensitivity than the results of BONEVAVI version 2.0.5 for prostate cancer (88%), breast cancer (86%) and lung cancer (82%), respectively. The study results demonstrate that the high sensitivity and specificity of the IFV system can provide assistance for image interpretation and generate prediction values for WBBS.

A Lightweight Convolutional Neural Network Architecture Applied for Bone Metastasis Classification in Nuclear Medicine: A Case Study on Prostate Cancer Patients

Bone metastasis is among the most frequent in diseases to patients suffering from metastatic cancer, such as breast or prostate cancer. A popular diagnostic method is bone scintigraphy where the whole body of the patient is scanned. However, hot spots that are presented in the scanned image can be misleading, making the accurate and reliable diagnosis of bone metastasis a challenge. Artificial intelligence can play a crucial role as a decision support tool to alleviate the burden of generating manual annotations on images and therefore prevent oversights by medical experts. So far, several state-of-the-art convolutional neural networks (CNN) have been employed to address bone metastasis diagnosis as a binary or multiclass classification problem achieving adequate accuracy (higher than 90%). However, due to their increased complexity (number of layers and free parameters), these networks are severely dependent on the number of available training images that are typically limited within the medical domain. Our study was dedicated to the use of a new deep learning architecture that overcomes the computational burden by using a convolutional neural network with a significantly lower number of floating-point operations (FLOPs) and free parameters. The proposed lightweight look-behind fully convolutional neural network was implemented and compared with several well-known powerful CNNs, such as ResNet50, VGG16, Inception V3, Xception, and MobileNet on an imaging dataset of moderate size (778 images from male subjects with prostate cancer). The results prove the superiority of the proposed methodology over the current state-of-the-art on identifying bone metastasis. The proposed methodology demonstrates a unique potential to revolutionize image-based diagnostics enabling new possibilities for enhanced cancer metastasis monitoring and treatment.