[en] The recent explosion of ‘big data’ has ushered in a new era of artificial intelligence (AI) algorithms in every sphere of technological activity, including medicine, and in particular radiology. However, the recent success of AI in certain flagship applications has, to some extent, masked decades-long advances in computational technology development for medical image analysis. In this article, we provide an overview of the history of AI methods for radiological image analysis in order to provide a context for the latest developments. We review the functioning, strengths and limitations of more classical methods as well as of the more recent deep learning techniques. We discuss the unique characteristics of medical data and medical science that set medicine apart from other technological domains in order to highlight not only the potential of AI in radiology but also the very real and often overlooked constraints that may limit the applicability of certain AI methods. Finally, we provide a comprehensive perspective on the potential impact of AI on radiology and on how to evaluate it not only from a technical point of view but also from a clinical one, so that patients can ultimately benefit from it.

Key Points

• Artificial intelligence (AI) research in medical imaging has a long history

• The functioning, strengths and limitations of more classical AI methods is reviewed, together with that of more recent deep learning methods.

• A perspective is provided on the potential impact of AI on radiology and on its evaluation from both technical and clinical points of view.

[en] The last few decades have witnessed tremendous technological developments in image-based biomarkers for tumor quantification and characterization. Initially limited to manual one- and two-dimensional size measurements, image biomarkers have evolved to harness developments not only in image acquisition technology but also in image processing and analysis algorithms. At the same time, clinical validation remains a major challenge for the vast majority of these novel techniques, and there is still a major gap between the latest technological developments and image biomarkers used in everyday clinical practice. Currently, the imaging biomarker field is attracting increasing attention not only because of the tremendous interest in cutting-edge therapeutic developments and personalized medicine but also because of the recent progress in the application of artificial intelligence (AI) algorithms to large-scale datasets. Thus, the goal of the present article is to review the current state of the art for image biomarkers and their use for characterization and predictive quantification of solid tumors. Beginning with an overview of validated imaging biomarkers in current clinical practice, we proceed to a review of AI-based methods for tumor characterization, such as radiomics-based approaches and deep learning.

Key Points

• Recent years have seen tremendous technological developments in image-based biomarkers for tumor quantification and characterization.

• Image-based biomarkers can be used on an ongoing basis, in a non-invasive (or mildly invasive) way, to monitor the development and progression of the disease or its response to therapy.

• We review the current state of the art for image biomarkers, as well as the recent developments in artificial intelligence (AI) algorithms for image processing and analysis.

[en] To examine the natural history of incidentally detected pancreatic cysts and whether a simplified MRI protocol without gadolinium is adequate for lesion follow-up. Over a 10-year period, 301-patients with asymptomatic pancreatic cysts underwent follow-up (45 months ± 30). The magnetic resonance imaging (MRI) protocol included axial, coronal T2-weighted images, MR cholangiopancreatographic and fat suppressed T1-weighted sequences before and after gadolinium. Three radiologists independently reviewed the initial MRI, the follow-up studies using first only unenhanced images, then secondly gadolinium-enhanced-sequences. Lesion changes during follow-up were recorded and the added value of gadolinium-enhanced sequences was determined by classifying the lesions into risk categories. Three hundred and one patients (1,174 cysts) constituted the study population. Only 35/301 patients (12 %) showed significant lesion change on follow-up. Using multivariate analysis the only independent factor of lesion growth (OR = 2.4; 95 % CI, 1.7-3.3; P < 0.001) and mural nodule development (OR = 1.9; 95 % CI, 1.1-3.4, P = 0.03) during follow-up was initial lesion size. No patient with a lesion initial size less than 2 cm developed cancer during follow-up. Intra-observer agreement with and without gadolinium enhancement ranged from 0.86 to 0.97. After consensus review of discordant cases, gadolinium-enhanced sequences demonstrated no added value. Most incidental pancreatic cystic lesions did not demonstrate change during follow-up. The addition of gadolinium-enhanced-sequences had no added-value for risk assignment on serial follow-up. (orig.)

[en] 

Background

Severe spontaneous soft tissue hematomas (SSTH) are usually treated with transcatheter arterial embolization (TAE) although only limited retrospective studies exist evaluating this treatment option. The aim of this study was to systematically assess the efficacy and safety of TAE for the management of SSTH.

Methods

Medline, EMBASE, PubMed and Cochrane Library were searched from inception to July 2017 using MeSH headings and a combination of keywords. Eligibility was restricted to original studies with patients suffering from SSTH treated with TAE. Patients with traumatic hematomas or who were treated with solely conservative or surgical management were excluded. For each publication, clinical success based on the control of the bleed, rebleeding rates and complications (including mortality) was collected, as well as technical details.

Results

Sixty-three studies met the inclusion criteria, with an aggregate total of 267 patients. Follow-up extended from 1 day to 10 years. Bleeding was mainly localized to the iliopsoas (n = 113/267, 42.3%) and anterior abdominal wall (n = 145/266, 54.7%). When information was available, 81.0% (n = 158/195) of patients were on anticoagulant therapy prior to the bleeding episode. Initial stabilization with control of the bleed was obtained in 93.1% (n = 242 patients, n = 60 studies). The most common embolic materials were coils (n = 129, 54.4%). Rebleeding was reported in 25 patients (9.4%). Only two embolization complications were reported (0.7%). The 30-day mortality was 22.7% (n = 42/1857).

Conclusion

TAE represents a safe and effective procedure in the management of SSTH. We present a management algorithm based on these data, but further studies are needed to address the knowledge gap.

[en] This study was conducted in order to evaluate a novel risk stratification model using dual-energy CT (DECT) texture analysis of head and neck squamous cell carcinoma (HNSCC) with machine learning to (1) predict associated cervical lymphadenopathy and (2) compare the accuracy of spectral versus single-energy (65 keV) texture evaluation for endpoint prediction. Eighty-seven patients with HNSCC were evaluated. Texture feature extraction was performed on virtual monochromatic images (VMIs) at 65 keV alone or different sets of multi-energy VMIs ranging from 40 to 140 keV, in addition to iodine material decomposition maps and other clinical information. Random forests (RF) models were constructed for outcome prediction with internal cross-validation in addition to the use of separate randomly selected training (70%) and testing (30%) sets. Accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were determined for predicting positive versus negative nodal status in the neck. Depending on the model used and subset of patients evaluated, an accuracy, sensitivity, specificity, PPV, and NPV of up to 88, 100, 67, 83, and 100%, respectively, could be achieved using multi-energy texture analysis. Texture evaluation of VMIs at 65 keV alone or in combination with only iodine maps had a much lower accuracy. Multi-energy DECT texture analysis of HNSCC is superior to texture analysis of 65 keV VMIs and iodine maps alone and can be used to predict cervical nodal metastases with relatively high accuracy, providing information not currently available by expert evaluation of the primary tumor alone.

[en] To test the performance of the Ovarian-Adnexal Reporting Data System (O-RADS) MRI in characterizing adnexal masses with cystic components and to test new specific MRI features related to cystic components to improve the ability of the O-RADS MRI score to stratify lesions according to their risk of malignancy. The EURopean ADnexal study (EURAD) database was retrospectively queried to identify adnexal masses with a cystic component. One junior and 13 radiologists independently reviewed cases blinded to the pathological diagnosis. For each lesion, the size of the whole lesion, morphological appearance, number of loculi, presence of a thickened wall, thickened septae, signal intensity of the cystic components on T1-weighted/T2-weighted/diffusion weighted, mean value of the apparent diffusion coefficient, and O-RADS MRI score were reported. Univariate and multivariate logistic regression analysis was performed to determine significant features to predict malignancy. The final cohort consisted of 585 patients with 779 pelvic masses who underwent pelvic MRI to characterize an adnexal mass(es). Histology served as the standard of reference. The diagnostic performance of the O-RADS MRI score was 0.944, ${}_{95\mathrm{\%<!--\; \%\; -->}}$CI [0.922-0.961]. Significant criteria associated with malignancy included an O-RADS MRI score ≥ 4, ADC${}_{mean}$ of cystic component > 1.69, number of loculi > 3, lesion size > 75 mm, the presence of a thick wall, and a low T1-weighted, a high T2-weighted, and a low diffusion-weighted signal intensity of the cystic component. Multivariate analysis demonstrated that an O-RADS MRI score ≥ combined with an ADC mean of the cystic component > 1.69, size > 75 mm, and low diffusion-weighted signal of the cystic component significantly improved the diagnostic performance up to 0.958, ${}_{95\mathrm{\%<!--\; \%\; -->}}$CI [0.938-0.973]. Cystic component analysis may improve the diagnosis performance of the O-RADS MRI score in adnexal cystic masses. O-RADS MRI score combined with specific cystic features (area under the receiving operating curve, AUROC = 0.958) improves the diagnostic performance of the O-RADS MRI score (AUROC = 0.944) for predicting malignancy in this cohort. Cystic features that improve the prediction of malignancy are ADC mean > 1.69 (OR = 7); number of loculi ≥ 3 (OR = 5.16); lesion size > 75 mm (OR = 4.40); the presence of a thick wall (OR = 3.59); a high T2-weighted signal intensity score 4 or 5 (OR = 3.30); a low T1-weighted signal intensity score 1, 2, or 3 (OR = 3.45); and a low diffusion-weighted signal intensity (OR = 2.12). An adnexal lesion with a cystic component rated O-RADS MRI score 4 and an ADC value of the cystic component < 1.69 associated with a low diffusion-weighted signal, has virtually a 0% risk of malignancy.

[en] There is a rich amount of quantitative information in spectral datasets generated from dual-energy CT (DECT). In this study, we compare the performance of texture analysis performed on multi-energy datasets to that of virtual monochromatic images (VMIs) at 65 keV only, using classification of the two most common benign parotid neoplasms as a testing paradigm. Forty-two patients with pathologically proven Warthin tumour (n = 25) or pleomorphic adenoma (n = 17) were evaluated. Texture analysis was performed on VMIs ranging from 40 to 140 keV in 5-keV increments (multi-energy analysis) or 65-keV VMIs only, which is typically considered equivalent to single-energy CT. Random forest (RF) models were constructed for outcome prediction using separate randomly selected training and testing sets or the entire patient set. Using multi-energy texture analysis, tumour classification in the independent testing set had accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of 92%, 86%, 100%, 100%, and 83%, compared to 75%, 57%, 100%, 100%, and 63%, respectively, for single-energy analysis. Multi-energy texture analysis demonstrates superior performance compared to single-energy texture analysis of VMIs at 65 keV for classification of benign parotid tumours. (orig.)

[en] In this work, we addressed fully automatic determination of tumor functional uptake from positron emission tomography (PET) images without relying on other image modalities or additional prior constraints, in the context of multicenter images with heterogeneous characteristics. In cervical cancer, an additional challenge is the location of the tumor uptake near or even stuck to the bladder. PET datasets of 232 patients from five institutions were exploited. To avoid unreliable manual delineations, the ground truth was generated with a semi-automated approach: a volume containing the tumor and excluding the bladder was first manually determined, then a well-validated, semi-automated approach relying on the Fuzzy locally Adaptive Bayesian (FLAB) algorithm was applied to generate the ground truth. Our model built on the U-Net architecture incorporates residual blocks with concurrent spatial squeeze and excitation modules, as well as learnable non-linear downsampling and upsampling blocks. Experiments relied on cross-validation (four institutions for training and validation, and the fifth for testing). The model achieved good Dice similarity coefficient (DSC) with little variability across institutions (0.80±0.03), with higher recall (0.90±0.05) than precision (0.75±0.05) and improved results over the standard U-Net (DSC 0.77±0.05, recall 0.87±0.02, precision 0.74±0.08). Both vastly outperformed a fixed threshold at 40% of SUVmax (DSC 0.33±0.15, recall 0.52±0.17, precision 0.30±0.16). In all cases, the model could determine the tumor uptake without including the bladder. Neither shape priors nor anatomical information was required to achieve efficient training. The proposed method could facilitate the deployment of a fully automated radiomics pipeline in such a challenging multicenter context.

[en] Highlights: • Size and location of resectable pancreatic adenocarcinomas are the main factors associated with resection status. • Presence of a contact between tumor and superior mesenteric/portal vein confluence was not associated with an R1 resection in multivariate analysis. • A simple CT score based on tumor size and location can help predicting an R1 resection. • Tumors involving the uncinate process or the neck and with maximal diameter ≥ 20 mm are at high-risk of an R1 resection. • Tumors located elsewhere (except the pancreaticoduodenal interface) and with a maximal diameter ≥ 30 mm are at high-risk of an R1 resection. - Abstract: Background: Negative-margin status is a prognostic indicator for long-term survival following curative intent resection for pancreatic adenocarcinoma. Patients at increased risk for positive-margin resections may benefit from neoadjuvant chemotherapy prior to resection. Methods: We retrospectively analyzed preoperative computed-tomography (CT) scans in 108 consecutive patients that underwent curative intent resection for a resectable pancreatic ductal adenocarcinoma from 2009 to 2016 in two academic hospitals. Two radiologists independently staged the tumor, including tumor location, size, and tumor-to-superior mesenteric/portal vein (SMV/PV) contact. Uni and multivariate analysis were performed to identify independent predictors of an R1 resection. Results: Twenty-nine patients had an R1 resection (26.9%). Tumor size, location, and presence of tumor-to-SMV/PV contact were significantly associated with an R1 resection. In multivariate analysis, the independent parameters associated with resection status were: tumor size (R² = 9.7), and tumor location (neck R² = 6.6; pancreaticoduodenal interface R² = 4.4; uncinate process R² = 4.1), but not tumor-to-SMV/PV contact (R² = 0.1, p = 0.7). A simple CT score was built based on tumor size and location. Patients with an R0 resectability score ≥3, i.e. patients with tumor size ≥30 mm (except when tumor location is at the pancreatico-duodenal interface) or patients with tumor size ≥20 mm AND tumor located in the uncinate process or neck, were at high-risk of an R1 resection (AUC, 0.82; sensitivity, 79%; specificity, 76%). This score also showed good diagnostic performances for predicting an R1 resection involving the medial resection margin only (AUC, 0.85). Conclusions: A simple score based on tumor location and size can accurately predict patients at high-risk of an R1 resection.

[en] To compare the diagnostic performance and inter-reader agreement of the CT-based v2019 versus v2005 Bosniak classification systems for risk stratification of cystic renal lesions (CRL). This retrospective study included adult patients with CRL identified on CT scan between 2005 and 2018. The reference standard was histopathology or a minimum 4-year imaging follow-up. The studies were reviewed independently by five readers (three senior, two junior), blinded to pathology results and imaging follow-up, who assigned Bosniak categories based on the 2005 and 2019 versions. Diagnostic performance of v2005 and v2019 Bosniak classifications for distinguishing benign from malignant lesions was calculated by dichotomizing CRL into the potential for ablative therapy (III-IV) or conservative management (I-IIF). Inter-reader agreement was calculated using Light's Kappa. One hundred thirty-nine patients with 149 CRL (33 malignant) were included. v2005 and v2019 Bosniak classifications achieved similar diagnostic performance with a sensitivity of 91% vs 91% and a specificity of 89% vs 88%, respectively. Inter-reader agreement for overall Bosniak category assignment was substantial for v2005 (κ = 0.78) and v2019 (κ = 0.75) between senior readers but decreased for v2019 when the Bosniak classification was dichotomized to conservative management (I-IIF) or ablative therapy (III-IV) (0.80 vs 0.71, respectively). For v2019, wall thickness was the morphological feature with the poorest inter-reader agreement (κ = 0.43 and 0.18 for senior and junior readers, respectively). No significant improvement in diagnostic performance and inter-reader agreement was shown between v2005 and v2019. The observed decrease in inter-reader agreement in v2019 when dichotomized according to management strategy may reflect the more stringent morphological criteria. Versions 2005 and 2019 Bosniak classifications achieved similar diagnostic performance, but the specificity of higher risk categories (III and IV) was not increased while one malignant lesion was downgraded to v2019 Bosniak category II (i.e., not subjected to further follow-up). Inter-reader agreement was similar between v2005 and v2019 but moderately decreased for v2019 when the Bosniak classification was dichotomized according to the potential need for ablative therapies (I-II-IIF vs III-IV).