Automatically Extracted Machine Learning Features from Preoperative CT to Early Predict Microvascular Invasion in HCC: The Role of the Zone of Transition (ZOT).

BACKGROUND: Microvascular invasion (MVI) is a consolidated predictor of hepatocellular carcinoma (HCC) recurrence after treatments. No reliable radiological imaging findings are available for preoperatively diagnosing MVI, despite some progresses of radiomic analysis. Furthermore, current MVI radiomic studies have not been designed for small HCC nodules, for which a plethora of treatments exists. This study aimed to identify radiomic MVI predictors in nodules ≤3.0 cm by analysing the zone of transition (ZOT), crossing tumour and peritumour, automatically detected to face the uncertainties of radiologist's tumour segmentation. METHODS: The study considered 117 patients imaged by contrast-enhanced computed tomography; 78 patients were finally enrolled in the radiomic analysis. Radiomic features were extracted from the tumour and the ZOT, detected using an adaptive procedure based on local image contrast variations. After data oversampling, a support vector machine classifier was developed and validated. Classifier performance was assessed using receiver operating characteristic (ROC) curve analysis and related metrics. RESULTS: The original 89 HCC nodules (32 MVI+ and 57 MVI-) became 169 (62 MVI+ and 107 MVI-) after oversampling. Of the four features within the signature, three are ZOT heterogeneity measures regarding both arterial and venous phases. On the test set (19MVI+ and 33MVI-), the classifier predicts MVI+ with area under the curve of 0.86 (95%CI (0.70-0.93), p∼10-5), sensitivity = 79% and specificity = 82%. The classifier showed negative and positive predictive values of 87% and 71%, respectively. CONCLUSIONS: The classifier showed the highest diagnostic performance in the literature, disclosing the role of ZOT heterogeneity in predicting the MVI+ status.

Human, All Too Human? An All-Around Appraisal of the "Artificial Intelligence Revolution" in Medical Imaging.

Artificial intelligence (AI) has seen dramatic growth over the past decade, evolving from a niche super specialty computer application into a powerful tool which has revolutionized many areas of our professional and daily lives, and the potential of which seems to be still largely untapped. The field of medicine and medical imaging, as one of its various specialties, has gained considerable benefit from AI, including improved diagnostic accuracy and the possibility of predicting individual patient outcomes and options of more personalized treatment. It should be noted that this process can actively support the ongoing development of advanced, highly specific treatment strategies (e.g., target therapies for cancer patients) while enabling faster workflow and more efficient use of healthcare resources. The potential advantages of AI over conventional methods have made it attractive for physicians and other healthcare stakeholders, raising much interest in both the research and the industry communities. However, the fast development of AI has unveiled its potential for disrupting the work of healthcare professionals, spawning concerns among radiologists that, in the future, AI may outperform them, thus damaging their reputations or putting their jobs at risk. Furthermore, this development has raised relevant psychological, ethical, and medico-legal issues which need to be addressed for AI to be considered fully capable of patient management. The aim of this review is to provide a brief, hopefully exhaustive, overview of the state of the art of AI systems regarding medical imaging, with a special focus on how AI and the entire healthcare environment should be prepared to accomplish the goal of a more advanced human-centered world.

A [68Ga]Ga-DOTANOC PET/CT Radiomic Model for Non-Invasive Prediction of Tumour Grade in Pancreatic Neuroendocrine Tumours.

Predicting grade 1 (G1) and 2 (G2) primary pancreatic neuroendocrine tumour (panNET) is crucial to foresee panNET clinical behaviour. Fifty-one patients with G1-G2 primary panNET demonstrated by pre-surgical [68Ga]Ga-DOTANOC PET/CT and diagnostic conventional imaging were grouped according to the tumour grade assessment method: histology on the whole excised primary lesion (HS) or biopsy (BS). First-order and second-order radiomic features (RFs) were computed from SUV maps for the whole tumour volume on HS. The RFs showing the lowest p-values and the highest area under the curve (AUC) were selected. Three radiomic models were assessed: A (trained on HS, validated on BS), B (trained on BS, validated on HS), and C (using the cross-validation on the whole dataset). The second-order normalized homogeneity and entropy was the most effective RFs couple predicting G2 and G1. The best performance was achieved by model A (test AUC = 0.90, sensitivity = 0.88, specificity = 0.89), followed by model C (median test AUC = 0.87, sensitivity = 0.83, specificity = 0.82). Model B performed worse. Using HS to train a radiomic model leads to the best prediction, although a "hybrid" (HS+BS) population performs better than biopsy-only. The non-invasive prediction of panNET grading may be especially useful in lesions not amenable to biopsy while [68Ga]Ga-DOTANOC heterogeneity might recommend FDG PET/CT.

Liver CT perfusion: which is the relevant delay that reduces radiation dose and maintains diagnostic accuracy?

OBJECTIVES: High radiation dose during CT perfusion (CTp) studies contributes to prevent CTp application in daily clinical practice. This work evaluates the consequences of scan delay on perfusion parameters and provides guidelines to help reducing the radiation dose by choosing the most appropriate delay. METHODS: Fifty-nine patients (34 men, 25 women; mean age 68 ± 12) with colorectal cancer, without underlying liver disease, underwent liver CTp, with the acquisition starting simultaneously with iodinated contrast agent injection. Blood flow (BF) and hepatic perfusion index (HPI) were computed on the acquired examinations and compared with those of the same examinations when a variable scan delay (τ) is introduced. Dose length product, CT dose index, and effective dose were also computed on original and delayed examinations. RESULTS: Altogether, three groups of delays (τ ≤ 4 s, 5 s ≤ τ ≤ 9 s, τ ≥ 10 s) were identified, yielding increasing radiation dose saving (RDS) (RDS ≤ 9.5%, 11.9% ≤ RDS ≤ 21.4%, RDS ≥ 23.8%) and decreasing perfusion accuracy (high (τ ≤ 4 s), medium (5 s ≤ τ ≤ 9 s), low (τ ≥ 10 s)). In particular, single-input and arterial BF and HPI were more insensitive to delay as regards the absolute variations (only 1 ml/min/100 g and 1%, respectively, for τ ≤ 9 s), than portal and total BF. CONCLUSION: Using delays lower than 4 s does not change perfusion accuracy and conveys unnecessary dose to patients. Conversely, starting the acquisition 9 s after contrast agent injection yields a RDS of about 21%, with no significant losses in perfusion accuracy. KEY POINTS: • Scan delays lower than 4 s do not alter perfusion accuracy and deliver an unnecessary radiation dose to patients. • Radiation dose delivered to patients can be reduced by 21.4% by introducing a 9-s scan delay, while keeping accurate perfusion values. • Using scan delays higher than 10 s, some perfusion parameters (portal and total BF) were inaccurate.

Multislice Analysis of Blood Flow Values in CT Perfusion Studies of Lung Cancer.

Objectives. Tumour heterogeneity represents a key issue in CT perfusion (CTp), where all studies are usually based on global mean or median values of perfusion maps, often computed on whole tumour. We sought to determine whether, and to what extent, such global values can be representative of tumour heterogeneity, with respect to single slices, and could be used for therapy assessment. Materials and Methods. Twelve patients with one primary non-small cell lung cancer lesion were enrolled in this study, for a total amount of 26 CTp examinations and 118 slices. Mean and median blood flow (BF) values, calculated voxel-based, were computed on each slice and the whole tumour. To measure functional heterogeneity, entropy was calculated on BF values as well. Results. Most of the slices were not represented by the global BF values computed on the whole tumour. In addition, there are a number of lesions having equivalent global BF values, but they are composed of slices having very different heterogeneity distributions, that is, entropy values. Conclusions. Global mean/median BF values of the single slices separately should be considered for clinical assessment, only if interpreted through entropy computed on BF values. The numerical equivalence between global BF values of different lesions may correspond to different clinical status, thus inducing possible errors in choice of therapy when considering global values only.

Effects of guided random sampling of TCCs on blood flow values in CT perfusion studies of lung tumors.

RATIONALE AND OBJECTIVES: Tissue perfusion is commonly used to evaluate lung tumor lesions through dynamic contrast-enhanced computed tomography (DCE-CT). The aim of this study was to improve the reliability of the blood flow (BF) maps by means of a guided sampling of the tissue time-concentration curves (TCCs). MATERIALS AND METHODS: Fourteen selected CT perfusion (CTp) examinations from different patients with lung lesions were considered, according to different degrees of motion compensation. For each examination, two regions of interest (ROIs) referring to the target lesion and the arterial input were manually segmented. To obtain the perfusion parameters, we computed the maximum slope of the Hill equation, describing the pharmacokinetics of the contrast agent, and the TCC was fitted for each voxel. A guided iterative approach based on the Random Sample Consensus method was used to detect and exclude samples arising from motion artifacts through the assessment of the confidence level of each single temporal sample of the TCC compared to the model. Removing these samples permits to refine the model fitting, thus exploiting more reliable data. Goodness-of-fit measures of the fitted TCCs to the original data (eg, root mean square error and correlation distance) were used to assess the reliability of the BF values, so as to preserve the functional structure of the resulting perfusion map. We devised a quantitative index, the local coefficient of variation (lCV), to measure the spatial coherence of perfusion maps, from local to regional and global resolution. The effectiveness of the algorithm was tested under three different degrees of motion yielded by as many alignment procedures. RESULTS: At pixel level, the proposed approach improved the reliability of BF values, quantitatively assessed through the correlation index. At ROI level, a comparative analysis emphasized how our approach "replaced" the noisy pixels, providing smoother parametric maps while preserving the main functional structure. Moreover, the implemented algorithm provides a more meaningful effect in correspondence of a higher motion degree. This was confirmed both quantitatively, using the lCV, and qualitatively, through visual inspection by expert radiologists. CONCLUSIONS: Perfusion maps achieved with the proposed approach can now be used as a valid tool supporting radiologists in DCE-CTp studies. This represents a step forward to clinical utilization of these studies for staging, prognosis, and monitoring values of therapeutic regimens.

Quantitative assessment of effects of motion compensation for liver and lung tumors in CT perfusion.

RATIONALE AND OBJECTIVES: To study the effects of four different rigid alignment approaches on both time-concentration curves (TCCs) and perfusion maps in computed tomography perfusion (CTp) studies of liver and lung tumors. MATERIALS AND METHODS: Eleven data sets in patients who were subjected to axial CTp after contrast agent administration were assessed. Each data set consists of four different sequences, according to the different rigid alignment configurations considered to compute blood flow perfusion maps: no alignment, translational, craniocaudal, and three dimensional (3D). The color maps were built on TCCs according to the maximum slope method. The effects of motion correction procedures on the reliability of TCCs and perfusion maps were assessed both quantitatively and visually. RESULTS: TCCs built after 3D alignments show the best indices as well as producing the most reliable maps. We show examinations in which the translational alignment only yields more accurate TCCs, but less reliable perfusion maps, than those achieved with no alignment. Furthermore, we show color maps with two different perfusion patterns, both considered reliable by radiologists, achieved with different motion correction approaches. CONCLUSIONS: The quantitative index we conceived allows relating quality of 3D alignment and reliability of perfusion maps. A better alignment does not necessarily yield more reliable perfusion values: color maps resulting from either alignment procedure must be critically assessed by radiologists. This achievement will hopefully represent a step forward for the clinical use of CTp studies for staging, prognosis, and monitoring values of therapeutic regimens.