AI applications to medical images: From machine learning to deep learning

PURPOSE

Artificial intelligence (AI) models are playing an increasing role in biomedical research and healthcare services. This review focuses on challenges points to be clarified about how to develop AI applications as clinical decision support systems in the real-world context.

METHODS

A narrative review has been performed including a critical assessment of articles published between 1989 and 2021 that guided challenging sections.

RESULTS

We first illustrate the architectural characteristics of machine learning (ML)/radiomics and deep learning (DL) approaches. For ML/radiomics, the phases of feature selection and of training, validation, and testing are described. DL models are presented as multi-layered artificial/convolutional neural networks, allowing us to directly process images. The data curation section includes technical steps such as image labelling, image annotation (with segmentation as a crucial step in radiomics), data harmonization (enabling compensation for differences in imaging protocols that typically generate noise in non-AI imaging studies) and federated learning. Thereafter, we dedicate specific sections to: sample size calculation, considering multiple testing in AI approaches; procedures for data augmentation to work with limited and unbalanced datasets; and the interpretability of AI models (the so-called black box issue). Pros and cons for choosing ML versus DL to implement AI applications to medical imaging are finally presented in a synoptic way.

CONCLUSIONS

Biomedicine and healthcare systems are one of the most important fields for AI applications and medical imaging is probably the most suitable and promising domain. Clarification of specific challenging points facilitates the development of such systems and their translation to clinical practice.

Validation of an optimized SPM procedure for FDG-PET in dementia diagnosis in a clinical setting

Diagnostic accuracy in FDG-PET imaging highly depends on the operating procedures. In this clinical study on dementia, we compared the diagnostic accuracy at a single-subject level of a) Clinical Scenarios, b) Standard FDG Images and c) Statistical Parametrical (SPM) Maps generated via a new optimized SPM procedure. We evaluated the added value of FDG-PET, either Standard FDG Images or SPM Maps, to Clinical Scenarios. In 88 patients with neurodegenerative diseases (Alzheimer's Disease—AD, Frontotemporal Lobar Degeneration—FTLD, Dementia with Lewy bodies—DLB and Mild Cognitive Impairment—MCI), 9 neuroimaging experts made a forced diagnostic decision on the basis of the evaluation of the three types of information. There was also the possibility of a decision of normality on the FDG-PET images. The clinical diagnosis confirmed at a long-term follow-up was used as the gold standard. SPM Maps showed higher sensitivity and specificity (96% and 84%), and better diagnostic positive (6.8) and negative (0.05) likelihood ratios compared to Clinical Scenarios and Standard FDG Images. SPM Maps increased diagnostic accuracy for differential diagnosis (AD vs. FTD; beta 1.414, p = 0.019). The AUC of the ROC curve was 0.67 for SPM Maps, 0.57 for Clinical Scenarios and 0.50 for Standard FDG Images. In the MCI group, SPM Maps showed the highest predictive prognostic value (mean LOC = 2.46), by identifying either normal brain metabolism (exclusionary role) or hypometabolic patterns typical of different neurodegenerative conditions.

•

Brain FDG-PET was evaluated with a new optimized SPM procedure in dementias.

•

We compared the diagnostic accuracy of clinical information, visual and SPM FDG-PET.

•

SPM had the best sensitivity (96%), specificity (84%) and positive and negative LR.

•

In an MCI subgroup, SPM had the highest predictive prognostic value.

Machine learning on brain MRI data for differential diagnosis of Parkinson's disease and Progressive Supranuclear Palsy

BACKGROUND

Supervised machine learning has been proposed as a revolutionary approach for identifying sensitive medical image biomarkers (or combination of them) allowing for automatic diagnosis of individual subjects. The aim of this work was to assess the feasibility of a supervised machine learning algorithm for the assisted diagnosis of patients with clinically diagnosed Parkinson's disease (PD) and Progressive Supranuclear Palsy (PSP).

METHOD

Morphological T1-weighted Magnetic Resonance Images (MRIs) of PD patients (28), PSP patients (28) and healthy control subjects (28) were used by a supervised machine learning algorithm based on the combination of Principal Components Analysis as feature extraction technique and on Support Vector Machines as classification algorithm. The algorithm was able to obtain voxel-based morphological biomarkers of PD and PSP.

RESULTS

The algorithm allowed individual diagnosis of PD versus controls, PSP versus controls and PSP versus PD with an Accuracy, Specificity and Sensitivity>90%. Voxels influencing classification between PD and PSP patients involved midbrain, pons, corpus callosum and thalamus, four critical regions known to be strongly involved in the pathophysiological mechanisms of PSP.

COMPARISON WITH EXISTING METHODS

Classification accuracy of individual PSP patients was consistent with previous manual morphological metrics and with other supervised machine learning application to MRI data, whereas accuracy in the detection of individual PD patients was significantly higher with our classification method.

CONCLUSIONS

The algorithm provides excellent discrimination of PD patients from PSP patients at an individual level, thus encouraging the application of computer-based diagnosis in clinical practice.

Radiomics and gene expression profile to characterise the disease and predict outcome in patients with lung cancer

Objective

The objectives of our study were to assess the association of radiomic and genomic data with histology and patient outcome in non-small cell lung cancer (NSCLC).

Methods

In this retrospective single-centre observational study, we selected 151 surgically treated patients with adenocarcinoma or squamous cell carcinoma who performed baseline [18F] FDG PET/CT. A subgroup of patients with cancer tissue samples at the Institutional Biobank (n = 74/151) was included in the genomic analysis. Features were extracted from both PET and CT images using an in-house tool. The genomic analysis included detection of genetic variants, fusion transcripts, and gene expression. Generalised linear model (GLM) and machine learning (ML) algorithms were used to predict histology and tumour recurrence.

Results

Standardised uptake value (SUV) and kurtosis (among the PET and CT radiomic features, respectively), and the expression of TP63, EPHA10, FBN2, and IL1RAP were associated with the histotype. No correlation was found between radiomic features/genomic data and relapse using GLM. The ML approach identified several radiomic/genomic rules to predict the histotype successfully. The ML approach showed a modest ability of PET radiomic features to predict relapse, while it identified a robust gene expression signature able to predict patient relapse correctly. The best-performing ML radiogenomic rule predicting the outcome resulted in an area under the curve (AUC) of 0.87.

Conclusions

Radiogenomic data may provide clinically relevant information in NSCLC patients regarding the histotype, aggressiveness, and progression. Gene expression analysis showed potential new biomarkers and targets valuable for patient management and treatment. The application of ML allows to increase the efficacy of radiogenomic analysis and provides novel insights into cancer biology.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00259-021-05371-7.

[18F]FDG PET/CT features for the molecular characterization of primary breast tumors

PURPOSE

The aim of this study was to evaluate the role of imaging features derived from [¹⁸F]FDG-PET/CT to provide in vivo characterization of breast cancer (BC).

METHODS

Images from 43 patients with a first diagnosis of BC were reviewed. Images were acquired before any treatment. Histological data were derived from pretreatment biopsy or surgical histological specimen; these included tumor type, grade, ER and PgR receptor status, lymphovascular invasion, Ki67 index, HER2 status, and molecular subtype. Standard parameters (SUV_mean, TLG, MTV) and advanced imaging features (histogram-based and shape and size features) were evaluated. Univariate analysis, hierarchical clustering analysis, and exact Fisher's test were used for statistical analysis of data. Imaging-derived metrics were reduced evaluating the mutual correlation within group of features as well as the mutual correlation between groups of features to form a signature.

RESULTS

A significant correlation was found between some advanced imaging features and the histological type. Different molecular subtypes were characterized by different values of two histogram-based features (median and energy). A significant association was observed between the imaging signature and luminal A and luminal B HER2 negative molecular subtype and also when considering luminal A, luminal B HER2-negative and HER2-positive groups. Similar results were found between the signature and all five molecular subtypes and also when considering the histological types of BC.

CONCLUSIONS

Our results suggest a complementary role of standard PET imaging parameters and advanced imaging features for the in vivo biological characterization of BC lesions.

Prospective comparison of MR with diffusion-weighted imaging, endoscopic ultrasound, MDCT and positron emission tomography-CT in the pre-operative staging of oesophageal cancer: results from a pilot study

OBJECTIVE

To compare the diagnostic performance of MR and diffusion-weighted imaging (DWI), multidetector CT, endoscopic ultrasonography (EUS) and ¹⁸F-FDG (fluorine-18 fludeoxyglucose) positron emission tomography CT (PET-CT) in the pre-operative locoregional staging of oesophageal cancer.

METHODS

18 patients with oesophageal or Siewert I tumour (9 directly treated with surgery and 9 addressed to chemo-/radiotherapy before) underwent 1.5-T MR and DWI, 64-channel multidetector CT, EUS and PET-CT before (n = 18) and also after neoadjuvant treatment (n = 9). All images were analysed and staged blindly by dedicated operators (seventh TNM edition). Two radiologists calculated independently the apparent diffusion coefficient from the first scan. Results were compared with histopathological findings. After the population had been divided according to local invasion (T1-T2 vs T3-T4) and nodal involvement (N0 vs N+), sensitivity, specificity, accuracy, positive- and negative-predictive values were calculated and compared. Quantitative measurements from DWI and PET-CT were also analysed.

RESULTS

For T staging, EUS showed the best sensitivity (100%), whereas MR showed the highest specificity (92%) and accuracy (83%). For N staging, MR and EUS showed the highest sensitivity (100%), but none of the techniques showed adequate results for specificity. Overall, MR showed the highest accuracy (66%) for N stage, although this was not significantly different to the other modalities. The apparent diffusion coefficient was different between surgery-only and chemo-/radiotherapy groups (1.90 vs 1.30 × 10^-3 mm² s^-1, respectively; p = 0.005)-optimal cut off for local invasion: 1.33 × 10^-3 mm² s^-1 (p = 0.05). Difference in standardized uptake value was also very close to conventional levels of statistical significance (8.81 vs 13.97 g cm^-3, respectively; p = 0.05)-optimal cut off: 7.97 g cm^-3 (p = 0.44).

CONCLUSION

In this pilot study, we have shown that MR with DWI could enrich the current pre-operative work-up for oesophageal cancer and could be used for T and N staging. However, larger studies will need to be carried out before introducing this technique in the standard diagnostic pathway, in order to understand if MR with DWI could change its management and replace more costly or invasive tests such as PET-CT or EUS. Advances in knowledge: This pilot study represents the first effort where the four techniques have been prospectively compared together for oesophageal cancer staging. The combination of MR and DWI could provide important, additional information for staging and initial treatment decision-making.

Response to chemotherapy in gastric adenocarcinoma with diffusion-weighted MRI and (18) F-FDG-PET/CT: correlation of apparent diffusion coefficient and partial volume corrected standardized uptake value with histological tumor regression grade

PURPOSE

To assess whether changes in diffusion-weighted MRI (DW-MRI) and (18) F-fluoro-2-deoxyglucose positron emission tomography/computed tomography ((18) F-FDG PET/CT), correlate with treatment response to neoadjuvant therapy (NT), as expressed by tumor regression grade (TRG), from locally advanced gastric adenocarcinoma (GA).

MATERIALS AND METHODS

Seventeen patients underwent both DW-MRI and (18) F-FDG-PET/CT scans before and after the end of NT. Apparent diffusion coefficient (ADC) and mean standardized uptake value (SUV) corrected for partial volume effect (PVC-SUVBW-mean ) were evaluated and compared with histopathological TRG.

RESULTS

Pre- and post-NT and percentage changes for ADC and PVC-SUVBW-mean were assessed. Post-NT ADC and ΔADC showed a significant inverse correlation with TRG (r = -0.71; P = 0.0011 and r = -0.78; P = 0.00020, respectively) and significant differences in their mean values were found between responders (TRG 1-2-3) and nonresponders (TRG 4-5) (P = 0.0009; P = 0.000082, respectively). No correlations with TRG were found for pre-NT ADC and for all PVC-SUVBW-mean values as well as between ΔADC and Δ PVC-SUVBW-mean .

CONCLUSION

DW-MRI seems more accurate than (18) F-FDG-PET/CT and ADC modifications may represent a reproducible tool to assess tumor response for GA.

Predictive value of pre-therapy (18)F-FDG PET/CT for the outcome of (18)F-FDG PET-guided radiotherapy in patients with head and neck cancer

PURPOSE

The aim of this study was to evaluate the predictive role of pre-therapy fluorodeoxyglucose (FDG) uptake parameters of primary tumour in head and neck cancer (HNC) patients undergoing intensity-modulated radiotherapy (IMRT) with simultaneous integrated boost (SIB) on FDG-positive volume-positron emission tomography (PET) gross tumour volume (PET-GTV).

METHODS

This retrospective study included 19 patients (15 men and 4 women, mean age 59.2 years, range 23-81 years) diagnosed with HNC between 2005 and 2011. Of 19 patients, 15 (79 %) had stage III-IV. All patients underwent FDG PET/CT before treatment. Metabolic indexes of primary tumour, including metabolic tumour volume (MTV), maximum and mean standardized uptake value (SUVmax, SUVmean) and total lesion glycolysis (TLG) were considered. Partial volume effect correction (PVC) was performed for SUVmean and TLG estimation. Correlations between PET/CT parameters and 2-year disease-free survival (DFS), local recurrence-free survival (LRFS) and distant metastasis-free survival (DMFS) were assessed. Median patient follow-up was 19.2 months (range 4-24 months).

RESULTS

MTV, TLG and PVC-TLG predicting patients' outcome with respect to all the considered local and distant disease control endpoints (LRFS, DMFS and DFS) were 32.4 cc, 469.8 g and 547.3 g, respectively. SUVmean and PVC-SUVmean cut-off values predictive of LRFS and DFS were 10.8 and 13.3, respectively. PVC was able to compensate errors up to 25 % in the primary HNC tumour uptake. Moreover, PVC enhanced the statistical significance of the results.

CONCLUSION

FDG PET/CT uptake parameters are predictors of patients' outcome and can potentially identify patients with higher risk of treatment failure that could benefit from more aggressive approaches. Application of PVC is recommended for accurate measurement of PET parameters.

New positron emission tomography derived parameters as predictive factors for recurrence in resected stage I non-small cell lung cancer

BACKGROUND

The recurrence rate for stage I non-small cell lung cancer is high, with 20-40% of patients that relapse after surgery. The aim of this study was to evaluate new F-18 fluorodeoxyglucose (FDG) positron emission tomography (PET) derived parameters, such as standardized uptake value index (SUVindex), metabolic tumor volume (MTV) and total lesion glycolysis (TLG), as predictive factors for recurrence in resected stage I non-small cell lung cancer.

METHODS

We retrospectively reviewed 99 resected stage I non-small cell lung cancer patients that were grouped by SUVindex, TLG and MTV above or below their median value. Disease free survival was evaluated as primary end point.

RESULTS

The 5-year overall survival and the 5-year disease free survival rates were 62% and 73%, respectively. The median SUVindex, MTL and TLG were 2.73, 2.95 and 9.61, respectively. Patients with low SUVindex, MTV and TLG were more likely to have smaller tumors (p ≤ 0.001). Univariate analysis demonstrated that SUVindex (p = 0.027), MTV (p = 0.014) and TLG (p = 0.006) were significantly related to recurrence showing a better predictive performance than SUVmax (p = 0.031). The 5-year disease free survival rates in patients with low and high SUVindex, MTV and TLG were 84% and 59%, 86% and 62% and 88% and 60%, respectively. The multivariate analysis showed that only TLG was an independent prognostic factor (p = 0.014) with a hazard ratio of 4.782.

CONCLUSION

Of the three PET-derived parameters evaluated, TLG seems to be the most accurate in stratifying surgically treated stage I non-small cell lung cancer patients according to their risk of recurrence.

Advanced Imaging Analysis in Prostate MRI: Building a Radiomic Signature to Predict Tumor Aggressiveness

Radiomics allows the extraction quantitative features from imaging, as imaging biomarkers of disease. The objective of this exploratory study is to implement a reproducible radiomic-pipeline for the extraction of a magnetic resonance imaging (MRI) signature for prostate cancer (PCa) aggressiveness. One hundred and two consecutive patients performing preoperative prostate multiparametric magnetic resonance imaging (mpMRI) and radical prostatectomy were enrolled. Multiparametric images, including T2-weighted (T2w), diffusion-weighted and dynamic contrast-enhanced images, were acquired at 1.5 T. Ninety-three imaging features (Ifs) were extracted from segmentation of index lesion. Ifs were ranked based on a stability rank and redundant Ifs were excluded. Using unsupervised hierarchical clustering, patients were grouped on the basis of similar radiomic patterns, whose association with Gleason Grade Group (GGG), extracapsular extension (ECE), and nodal involvement (pN) was tested. Signatures composed by IFs from T2w-images and Apparent Diffusion Coefficient (ADC) maps were tested for the prediction of GGG, ECE, and pN. T2w radiomic pattern was associated with pN, ECE, and GGG (p = 0.027, 0.05, 0.03) and ADC radiomic pattern was associated with GGG (p = 0.004). The best performance was reached by the signature combing IFs from multiparametric images (0.88, 0.89, and 0.84 accuracy for GGG, pN, and ECE). A reliable multiparametric MRI radiomic signature was extracted, potentially able to predict PCa aggressiveness, to be further validated on an independent sample.

Statistical Voxel-Based Methods and [18F]FDG PET Brain Imaging: Frontiers for the Diagnosis of AD

Recommended guidelines for the diagnosis of dementia due to Alzheimer's Disease (AD) were revised in recent years, including Positron Emission Tomography (PET) as an in-vivo diagnostic imaging technique for the diagnosis of neurodegeneration. In particular PET, using 18Ffluorodeoxiglucouse ([18F]FDG), is able to detect very early changes of glucose consumption at the synaptic level, enabling to support both early and differential diagnosis of AD. In standard clinical practice, interpretation of [18F] FDG-PET images is usually achieved through qualitative assessment. Visual inspection although only reveals information visible at human eyes resolution, while information at a higher resolution is missed. Furthermore, qualitative assessment depends on the degree of expertise of the clinician, preventing from the definition of accurate and standardized imaging biomarkers. Automated and computerized image processing methods have been proposed to support the in-vivo assessment of brain PET studies. In particular, objective statistical image analyses, enabling the comparison of one patient's images to a group of control images have been shown to carry important advantages for detecting significant metabolic changes, including the availability of more objective, cross-center reliable metrics and the detectability of brain subtle functional changes, as occurring in prodromal AD. The purpose of the current review is to provide a systematic overview encompassing the frontiers recently reached by quantitative approaches for the statistical analysis of PET brain images in the study of AD, with a particular focus on Statistical Parametric Mapping. Main achievements, e.g. in terms of standardized biomarkers of AD as well as of sensitivity and specificity, will be discussed.

Combination of gene expression and genome copy number alteration has a prognostic value for breast cancer

Specific genome copy number alterations, such as deletions and amplifications are an important factor in tumor development and progression, and are also associated with changes in gene expression. By combining analyses of gene expression and genome copy number we identified genes as candidate biomarkers of BC which were validated as prognostic factors of the disease progression. These results suggest that the proposed combined approach may become a valuable method for BC prognosis.

Acute Inescapable Stress Rapidly Increases Synaptic Energy Metabolism in Prefrontal Cortex and Alters Working Memory Performance

Brain energy metabolism actively regulates synaptic transmission and activity. We have previously shown that acute footshock (FS)-stress induces fast and long-lasting functional and morphological changes at excitatory synapses in prefrontal cortex (PFC). Here, we asked whether FS-stress increased energy metabolism in PFC, and modified related cognitive functions. Using positron emission tomography (PET), we found that FS-stress induced a redistribution of glucose metabolism in the brain, with relative decrease of [18F]FDG uptake in ventro-caudal regions and increase in dorso-rostral ones. Absolute [18F]FDG uptake was inversely correlated with serum corticosterone. Increased specific hexokinase activity was also measured in purified PFC synaptosomes (but not in total extract) of FS-stressed rats, which positively correlated with 2-Deoxy [3H] glucose uptake by synaptosomes. In line with increased synaptic energy demand, using an electron microscopy-based stereological approach, we found that acute stress induced a redistribution of mitochondria at excitatory synapses, together with an increase in their volume. The fast functional and metabolic activation of PFC induced by acute stress, was accompanied by rapid and sustained alterations of working memory performance in delayed response to T-maze test. Taken together, the present data suggest that acute stress increases energy consumption at PFC synaptic terminals and alters working memory.

Biomarkers from in vivo molecular imaging of breast cancer: pretreatment 18F-FDG PET predicts patient prognosis, and pretreatment DWI-MR predicts response to neoadjuvant chemotherapy

OBJECTIVE

Human cancers display intra-tumor phenotypic heterogeneity and recent research has focused on developing image processing methods extracting imaging descriptors to characterize this heterogeneity. This work assesses the role of pretreatment 18F-FDG PET and DWI-MR with respect to the prognosis and prediction of neoadjuvant chemotherapy (NAC) outcomes when image features are used to characterize primitive lesions from breast cancer (BC).

MATERIALS AND METHODS

A retrospective protocol included 38 adult women with biopsy-proven BC. Patients underwent a pre-therapy 18F-FDG PET/CT whole-body study and a pre-therapy breast multi-parametric MR study. Patients were then referred for NAC treatment and then for surgical resection, with an evaluation of the therapy response. Segmentation methods were developed in order to identify functional volumes both on 18F-FDG PET images and ADC maps. Macroscopic and histogram features were extracted from the defined functional volumes.

RESULTS

Our work demonstrates that macroscopic and histogram features from 18F-FDG PET are able to biologically characterize primitive BC, and define the prognosis. In addition, histogram features from ADC maps are able to predict the response to NAC.

CONCLUSION

Our work suggests that pre-treatment 18F-FDG PET and pre-treatment DWI-MR provide useful complementary information for biological characterization and NAC response prediction in BC.

The impact of different 18FDG PET healthy subject scans for comparison with single patient in SPM analysis

BACKGROUND

Statistical Parametric Mapping (SPM) has been applied for single-subject evaluation of [18F]FDG uptake in Alzheimer Disease (AD). In a single-subject framework, the patient is compared to a dataset of [18F]FDG PET images from healthy subjects (HS) evaluating brain metabolic abnormalities. No studies exist that assess the effects on SPM analysis of HS [18F]FDG PET datasets acquired from different subjects and using different PET scanners including the same or different PET scanners than those used for patients. This work aims to elucidate this issue from a methodological perspective.

METHODS

We considered six different [18F]FDG PET datasets, from different HS populations, acquired by different PET scanners. We applied SPM5 procedures for single-subject comparison with each of the six HS datasets in 10 probable AD patients showing the typical [18F]FDG pattern. We also implemented the same comparison in 3 probable AD patients and in 7 patients with a clinical diagnosis of Mild Cognitive Impariment (MCI), showing subtle changes on visual inspection of [18F]FDG distribution.

RESULTS

Considering the 10 patients with the typical [18F]FDG pattern, the results were comparable for all the SPM maps. In the 3 probable AD patients with subtle changes in [18F]FDG distribution, no significant AD pattern emerged when a small number (<20) of HS was used, whereas a significant AD pattern appeared when a large (>50) HS image set was used. In the 7 considered MCI patients the use of a large (>50) HS image set allowed to assess significant hypometabolic patterns related to a probable neurodegenerative pathology.

CONCLUSIONS

The use of large HS datasets of PET scans (>50) is recommended for single-subject SPM analysis. On condition that appropriate preprocessing steps are provided, large HS datasets can include HS images acquired with different PET systems, not including images from the same scanner of that used for patients.

Targeted radionuclide therapy: frontiers in theranostics

The concept of targeted radionuclide therapy (TRT) relies on the use of injected nuclear medicine as treating agents, targeted at the cellular or molecular level. The growth of the interest in TRT was stimulated by the advances in radionuclide production and labeling as well as by the improvement in the knowledge of appropriate and specific molecular targets. In recent years, different studies on TRT were focused on the evaluation of radionuclide compounds able to combine imaging of the disease with TRT, in a theranostic approach. This approach is of particular interest towards the personalization of treatments, allowing both the baseline characterization of oncological pathologies and treatment optimization by correct dosimetric calculation as well as therapy monitoring. This paper presents a review of recent literature on TRT, with a particular focus on clinical applications promoting such a theranostic approach, showing the impact of the synergy of diagnostic imaging and therapeutics.

In Silico Approach for the Definition of radiomiRNomic Signatures for Breast Cancer Differential Diagnosis

Personalized medicine relies on the integration and consideration of specific characteristics of the patient, such as tumor phenotypic and genotypic profiling.

BACKGROUND

Radiogenomics aim to integrate phenotypes from tumor imaging data with genomic data to discover genetic mechanisms underlying tumor development and phenotype.

METHODS

We describe a computational approach that correlates phenotype from magnetic resonance imaging (MRI) of breast cancer (BC) lesions with microRNAs (miRNAs), mRNAs, and regulatory networks, developing a radiomiRNomic map. We validated our approach to the relationships between MRI and miRNA expression data derived from BC patients. We obtained 16 radiomic features quantifying the tumor phenotype. We integrated the features with miRNAs regulating a network of pathways specific for a distinct BC subtype.

RESULTS

We found six miRNAs correlated with imaging features in Luminal A (miR-1537, -205, -335, -337, -452, and -99a), seven miRNAs (miR-142, -155, -190, -190b, -1910, -3617, and -429) in HER2+, and two miRNAs (miR-135b and -365-2) in Basal subtype. We demonstrate that the combination of correlated miRNAs and imaging features have better classification power of Luminal A versus the different BC subtypes than using miRNAs or imaging alone.

CONCLUSION

Our computational approach could be used to identify new radiomiRNomic profiles of multi-omics biomarkers for BC differential diagnosis and prognosis.

A tri-modal tissue-equivalent anthropomorphic phantom for PET, CT and multi-parametric MRI radiomics

PURPOSE

Radiomics has emerged as an advanced image processing methodology to define quantitative imaging biomarkers for prognosis and prediction of treatment response and outcome. The development of quantitative imaging biomarkers requires careful analysis to define their accuracy, stability and reproducibility through phantom measurements. Few efforts were devoted to develop realistic anthropomorphic phantoms. In this work, we developed a multimodality image phantom suitable for PET, CT and multiparametric MRI imaging.

METHODS

A tissue-equivalent gel-based mixture was designed and tested for compatibility with different imaging modalities. Calibration measurements allowed to assess gel composition to simulate PET, CT and MRI contrasts of oncological lesions. The characterized gel mixture was used to create realistic synthetic lesions (e.g. lesions with irregular shape and non-uniform image contrast), to be inserted in a standard anthropomorphic phantom. In order to show phantom usefulness, issues related to accuracy, stability and reproducibility of radiomic biomarkers were addressed as proofs-of-concept.

RESULTS

The procedure for gel preparation was straightforward and the characterized gel mixture allowed to mime simultaneously oncological lesion contrast in CT, PET and MRI imaging. Proofs-of-concept studies suggested that phantom measurements can be customized for specific clinical situations and radiomic protocols.

CONCLUSIONS

We developed a strategy to manufacture an anthropomorphic, tissue-equivalent, multimodal phantom to be customized on specific radiomics protocols, for addressing specific methodological issues both in mono and multicentric studies.

Validity of cingulate–precuneus–temporo-parietal hypometabolism for single-subject diagnosis of biomarker-proven atypical variants of Alzheimer’s Disease

The aim of our study was to establish empirically to what extent reduced glucose uptake in the precuneus, posterior cingulate and/or temporo-parietal cortex (PCTP), which is thought to indicate brain amyloidosis in patients with dementia or MCI due to Alzheimer’s Disease (AD), permits to distinguish amyloid-positive from amyloid-negative patients with non-classical AD phenotypes at the single-case level. We enrolled 127 neurodegenerative patients with cognitive impairment and a positive (n. 63) or negative (n. 64) amyloid marker (cerebrospinal fluid or amy-PET). Three rating methods of FDG-PET scan were applied: purely qualitative visual interpretation of uptake images (VIUI), and visual reading assisted by a semi-automated and semi-quantitative tool: INLAB, provided by the Italian National Research Council, or Cortex ID Suite, marketed by GE Healthcare. Fourteen scans (11.0%) patients remained unclassified by VIUI or INLAB procedures, therefore, validity values were computed on the remaining 113 cases. The three rating approaches showed good total accuracy (77–78%), good to optimal sensitivity (81–93%), but poorer specificity (62–75%). VIUI showed the highest sensitivity and the lowest specificity, and also the highest proportion of unclassified cases. Cases with asymmetric temporo-parietal hypometabolism and a progressive aphasia or corticobasal clinical profile, in particular, tended to be rated as AD-like, even if biomarkers indicated non-amyloid pathology. Our findings provide formal support to the value of PCTP hypometabolism for single-level diagnosis of amyloid pathophysiology in atypical AD, but also highlight the risk of qualitative assessment to misclassify patients with non-AD PPA or CBS underpinned by asymmetric temporo-parietal hypometabolism.