Convolutional neural network (CNN), a class of artificial neural networks that has become dominant in various computer vision tasks, is attracting interest across a variety of domains, including ...radiology. CNN is designed to automatically and adaptively learn spatial hierarchies of features through backpropagation by using multiple building blocks, such as convolution layers, pooling layers, and fully connected layers. This review article offers a perspective on the basic concepts of CNN and its application to various radiological tasks, and discusses its challenges and future directions in the field of radiology. Two challenges in applying CNN to radiological tasks, small dataset and overfitting, will also be covered in this article, as well as techniques to minimize them. Being familiar with the concepts and advantages, as well as limitations, of CNN is essential to leverage its potential in diagnostic radiology, with the goal of augmenting the performance of radiologists and improving patient care.
Key Points
• Convolutional neural network is a class of deep learning methods which has become dominant in various computer vision tasks and is attracting interest across a variety of domains, including radiology.
• Convolutional neural network is composed of multiple building blocks, such as convolution layers, pooling layers, and fully connected layers, and is designed to automatically and adaptively learn spatial hierarchies of features through a backpropagation algorithm.
• Familiarity with the concepts and advantages, as well as limitations, of convolutional neural network is essential to leverage its potential to improve radiologist performance and, eventually, patient care.
Objective
To analyze the natural history of small asymptomatic pancreatic neuroendocrine tumors (PanNET) and to present a matched comparison between groups who underwent either initial observation or ...resection. Management approach for small PanNET is uncertain.
Methods
Incidentally discovered, sporadic, small (<3 cm), stage I–II PanNET were analyzed retrospectively between 1993 and 2013. Diagnosis was determined either by pathology or imaging characteristics. Intention-to-treat analysis was applied.
Results
A total of 464 patients were reviewed. Observation was recommended for 104 patients (observation group), and these patients were matched to 77 patients in the resection group based on tumor size at initial imaging. The observation group was significantly older (median 63 vs. 59 years,
p
= 0.04) and tended towards shorter follow-up (44 vs. 57 months,
p
= 0.06). Within the observation group, 26 of the 104 patients (25 %) underwent subsequent tumor resection after a median observation interval of 30 months (range 7–135). At the time of last follow-up of the observation group, the median tumor size had not changed (1.2 cm,
p
= 0.7), and no patient had developed evidence of metastases. Within the resection group, low-grade (G1) pathology was recorded in 72 (95 %) tumors and 5 (6 %) developed a recurrence, which occurred after a median of 5.1 (range 2.9–8.1) years. No patient in either group died from disease. Death from other causes occurred in 11 of 181 (6 %) patients.
Conclusions
In this study, no patient who was initially observed developed metastases or died from disease after a median follow-up of 44 months. Observation for stable, small, incidentally discovered PanNET is reasonable in selected patients.
Background
Reports show that FOLFIRINOX therapy for pancreatic ductal adenocarcinoma (PDAC) results in objective response rates two to threefold higher than those of other regimens. This study aimed ...to assess response and resection rates for locally unresectable (stage 3) patients initially treated with induction FOLFIRINOX.
Methods
The institutional cancer database was queried for patients treated with induction FOLFIRINOX therapy between 2010 and 2013. Patients were included in the study if they were treated at the authors’ institution for stage 3 PDAC (locally unresectable) that had been adjudicated at a weekly multidisciplinary tumor board.
Results
The study identified 101 patients. The median age was 64 years (range 37–81 years), and the median follow-up period was 12 months (range 3–37 months). The patients received a median of six cycles (range 1–20 cycles) of induction FOLFIRINOX. No grade 4 or 5 toxicity was recorded. At the initial restaging (median of 3 months after diagnosis), 23 patients (23 %) had developed distant metastases, 15 patients (15 %) had undergone resection, and 63 patients (63 %) had proceeded to chemoradiation. In the group of 63 patients who had proceeded to chemoradiation (median of 9 months after diagnosis), an additional 16 patients (16 %) had undergone resection, and 5 patients (5 %) had developed metastases. A partial radiographic response was observed in 29 % of all the patients, which was associated with ability to perform resection (
p
= 0.004). The median overall survival time was 11 months for the group that progressed with FOLFIRINOX and 26 months for the group that did not progress.
Conclusion
Nearly one third of the patients who had been initially identified as having stage 3 pancreatic carcinoma and had been treated with FOLFIRINOX responded radiographically and underwent tumor resection.
The purpose of the current study was to develop a deep learning technique called Golden‐angle RAdial Sparse Parallel Network (GRASPnet) for fast reconstruction of dynamic contrast‐enhanced 4D MRI ...acquired with golden‐angle radial k‐space trajectories. GRASPnet operates in the image‐time space and does not use explicit data consistency to minimize the reconstruction time. Three different network architectures were developed: (1) GRASPnet‐2D: 2D convolutional kernels (x,y) and coil and contrast dimensions collapsed into a single combined dimension; (2) GRASPnet‐3D: 3D kernels (x,y,t); and (3) GRASPnet‐2D + time: two 3D kernels to first exploit spatial correlations (x,y,1) followed by temporal correlations (1,1,t). The networks were trained using iterative GRASP reconstruction as the reference. Free‐breathing 3D abdominal imaging with contrast injection was performed on 33 patients with liver lesions using a T1‐weighted golden‐angle stack‐of‐stars pulse sequence. Ten datasets were used for testing. The three GRASPnet architectures were compared with iterative GRASP results using quantitative and qualitative analysis, including impressions from two body radiologists. The three GRASPnet techniques reduced the reconstruction time to about 13 s with similar results with respect to iterative GRASP. Among the GRASPnet techniques, GRASPnet‐2D + time compared favorably in the quantitative analysis. Spatiotemporal deep learning enables reconstruction of dynamic 4D contrast‐enhanced images in a few seconds, which would facilitate translation to clinical practice of compressed sensing methods that are currently limited by long reconstruction times.
GRASPnet proposes a novel deep learning framework in the image domain to reconstruct motion‐sorted time‐resolved dynamic contrast‐enhanced MR images from aliased radial undersampled images in a few seconds. Qualitative and quantitative comparison of the GRASPnet results with those of the existing method, along with the radiologists’ impressions, shows good agreement between them, while GRASPnet enables rapid reconstruction allowing clinical translation.
BRCA-associated cancers have increased sensitivity to poly(ADP-ribose) polymerase inhibitors (PARPis). This single arm, non-randomised, multicentre phase II trial evaluated the response rate of ...veliparib in patients with previously treated BRCA1/2- or PALB2-mutant pancreatic adenocarcinoma (PDAC).
Patients with stage III/IV PDAC and known germline BRCA1/2 or PALB2 mutation, 1–2 lines of treatment, Eastern Cooperative Oncology Group 0–2, were enrolled. Veliparib was dosed at a volume of 300 mg twice-daily (N = 3), then 400 mg twice-daily (N = 15) days 1–28. The primary end-point was to determine the response rate of veliparib; secondary end-points included progression-free survival (PFS), duration of response, overall survival (OS) and safety.
Sixteen patients were enrolled; male N = 8 (50%). Median age was 52 years (range 43–77). Five (31%) had a BRCA1 and 11 (69%) had a BRCA2 mutation. Fourteen (88%) patients had received prior platinum-based therapy. No confirmed partial responses (PRs) were seen: one (6%) unconfirmed PR was observed at 4 months with disease progression (PD) at 6 months; four (25%) had stable disease (SD), whereas 11 (69%) had PD as best response including one with clinical PD. Median PFS was 1.7 months (95% confidence interval CI 1.57–1.83) and median OS was 3.1 months (95% CI 1.9–4.1). Six (38%) patients had grade III toxicity, including fatigue (N = 3), haematology (N = 2) and nausea (N = 1).
Veliparib was well tolerated, but no confirmed response was observed although four (25%) patients remained on study with SD for ≥ 4 months. Additional strategies in this population are needed, and ongoing trials are evaluating PARPis combined with chemotherapy (NCT01585805) and as a maintenance strategy (NCT02184195).
•Veliparib has modest activity in previously treated germline BRCA-mutated pancreas cancer exposed to prior platinum therapy.•Poly(ADP)-ribose) polymerase (PARP) inhibition in pancreas cancer may be best evaluated in a platinum-sensitive disease setting.•There may be intrinsic differences between PARP inhibitors that explain therapeutic outcomes.
Purpose
To evaluate the prevalence of major and ancillary imaging features from liver imaging reporting and data systems (LI-RADS) version 2014 and their interreader agreement when comparing ...hepatocellular carcinoma (HCC) to intrahepatic cholangiocarcinoma (ICC) and combined tumor (cHCC-CC).
Methods
The Institutional Review Board approved this HIPAA-compliant retrospective study and waived the requirement for patients’ informed consent. Patients with resected HCC (
n
= 51), ICC (
n
= 40), and cHCC-CC (
n
= 11) and available pre-operative contrast-enhanced MRI were included from 2000 to 2015. Imaging features and final LI-RADS category were evaluated by four radiologists. Imaging features were compared by Fisher’s exact test and interreader agreements were assessed by
κ
statistics.
Results
None of the features were unique to either HCC or non-HCC. Imaging features that were significantly more common among HCC compared to ICC and cHCC-CC included washout (76%–78% vs. 10%–35%,
p
< 0.001), capsule (55%–71% vs. 16%–49%,
p
< 0.05), and intralesional fat (27%–52% vs. 2%–12%,
p
< 0.002). Features that were more common among ICC and cHCC-CC included peripheral arterial phase hyperenhancement (40%–64% vs. 10%–14%,
p
< 0.001) and progressive central enhancement (65%–82% vs. 14%–25%,
p
< 0.001). The interreader agreement was moderate for each of these imaging features (
κ
= 0.41–0.55). Moderate agreement was also achieved in the assignment of LR-M (
κ
= 0.53), with an overall sensitivity and specificity for non-HCC malignancy of 86.3% and 78.4%, respectively.
Conclusion
HCC and non-HCC show significant differences in the prevalence of imaging features defined by LI-RADS, and are identified by radiologists with moderate interreader agreement. Using LI-RADS, radiologists also achieved moderate interreader agreement in the assignment of the LR-M category.
Comparative protein structure prediction is limited
mostly by the errors in alignment and loop modeling. We
describe here a new automated modeling technique that significantly
improves the accuracy ...of loop predictions in protein structures.
The positions of all nonhydrogen atoms of the loop are
optimized in a fixed environment with respect to a pseudo
energy function. The energy is a sum of many spatial restraints
that include the bond length, bond angle, and improper
dihedral angle terms from the CHARMM-22 force field, statistical
preferences for the main-chain and side-chain dihedral
angles, and statistical preferences for nonbonded atomic
contacts that depend on the two atom types, their distance
through space, and separation in sequence. The energy function
is optimized with the method of conjugate gradients combined
with molecular dynamics and simulated annealing. Typically,
the predicted loop conformation corresponds to the lowest
energy conformation among 500 independent optimizations.
Predictions were made for 40 loops of known structure at
each length from 1 to 14 residues. The accuracy of loop
predictions is evaluated as a function of thoroughness
of conformational sampling, loop length, and structural
properties of native loops. When accuracy is measured by
local superposition of the model on the native loop, 100,
90, and 30% of 4-, 8-, and 12-residue loop predictions,
respectively, had <2 Å RMSD error for the mainchain
N, Cα, C, and O atoms; the average accuracies
were 0.59 ± 0.05, 1.16 ± 0.10, and 2.61 ±
0.16 Å, respectively. To simulate real comparative
modeling problems, the method was also evaluated by predicting
loops of known structure in only approximately correct
environments with errors typical of comparative modeling
without misalignment. When the RMSD distortion of the main-chain
stem atoms is 2.5 Å, the average loop prediction
error increased by 180, 25, and 3% for 4-, 8-, and 12-residue
loops, respectively. The accuracy of the lowest energy
prediction for a given loop can be estimated from the structural
variability among a number of low energy predictions. The
relative value of the present method is gauged by (1) comparing
it with one of the most successful previously described
methods, and (2) describing its accuracy in recent blind
predictions of protein structure. Finally, it is shown
that the average accuracy of prediction is limited primarily
by the accuracy of the energy function rather than by the
extent of conformational sampling.
To examine outcomes of incidental liver lesions on baseline breast magnetic resonance imaging (MRI) that were further evaluated with dedicated abdominal imaging.
Consecutive breast MRI reports from ...2011 to 2016 were retrospectively reviewed to identify incidental liver lesions. Only patients without prior breast MRI, without prior abdominal imaging, and with subsequent abdominal imaging were included. Patient demographics, breast MRI indication, and final liver lesion diagnosis were recorded.
Of 131 women (mean age 53.8 years), 94/131 (71.8%) underwent breast MRI for extent of disease evaluation, 25/131 (19.1%) for high-risk screening, 11/131 (8.4%) for implant evaluation, and 1/131 (0.8%) for problem-solving. Of 131 liver lesions (6–80 mm), 117/131 (89.3%) were deemed benign on subsequent abdominal imaging; 10/131 (7.6%) probably benign; and 4/131 lesions (3.1%) were confirmed breast cancer metastases. Metastatic liver lesions identified on breast MRI were more likely for women with a current diagnosis of breast cancer than for women without a current diagnosis of breast cancer: 4.3% vs 0%. Similarly, metastatic liver lesions identified on breast MRI were more likely for those with a higher prognostic stage (2 or 3) vs a lower prognostic stage (0 or 1) or no current breast cancer: 11.1% vs 0%.
Baseline breast MRIs showing incidental liver lesions showed unsuspected liver metastases only in women with a current diagnosis of clinical stage 2 or 3 breast cancer. This suggests breast MRI indication and clinical staging of current breast cancer, if present, can help plan management and decisions to obtain follow-up of liver lesions.
•Incidental liver lesions are often encountered on baseline breast MRI.•Liver MRI is often used to characterize incidental lesions found on breast MRI.•Liver lesions are unlikely malignant in women without a known current cancer.•Liver lesions are more likely metastatic in higher stage breast cancer patients.
To perform a systematic review and meta-analysis to determine risk factors for hypervascularization in hepatobiliary phase (HBP) hypointense nodules without arterial phase hyperenhancement (APHE) in ...patients with hepatocellular carcinoma (HCC).
Pubmed and EMBASE databases were searched up to May 7, 2020. Studies which evaluated radiologic and clinical risk factors for hypervascularization in HBP hypointense nodules without APHE were included. Hazard ratios were meta-analytically pooled using random-effects model. Methodological quality of included studies was assessed using Quality in Prognostic Studies (QUIPS) tool.
Sixteen studies with 934 patients were included. HBP hypointense nodules without APHE with baseline size greater than 10 mm, T2 hyperintensity, and restricted diffusion showed risk for hypervascularization with pooled HRs of 2.95 (95% confidence interval CI, 1.94-4.20), 4.21 (95% CI, 1.15-15.40), 5.83 (95% CI, 1.42-23.95), respectively. Previous HCC history contributed to hypervascularization of the nodules with hazard ratio of 2.06 (95% CI, 1.23-3.44). T1 hyperintensity, intralesional fat, Child-Pugh Class B, sex, alfa-fetoprotein, hepatitis B or C infection were not significant risk factors for hypervascularization (p ≥0.05). Study quality was generally moderate.
HBP hypointense nodules without APHE on gadoxetic acid-enhanced MRI with baseline size greater than 10 mm, T2 hyperintensity, restricted diffusion and previous hepatocellular carcinoma history pose higher risk for hypervascularization. Proper patient management in patients with HBP hypointense nodules without APHE on gadoxetic acid-enhanced MRI may need to be tailored according to these risk factors.
Lutetium-177 (177Lu)-DOTATATE received FDA approval in 2018 to treat somatostatin receptor-positive gastroenteropancreatic neuroendocrine tumors (NETs). Little data are available on response and ...outcomes for well-differentiated (WD) high-grade (HG) NETs treated with 177Lu-DOTATATE.
Patients with WD HG NETs treated with 177Lu-DOTATATE at MSK from 2018 to 2020 were identified. Demographics, response (RECIST 1.1), and progression-free survival (PFS) were determined. Next-generation sequencing (NGS) was performed in the archival tumor.
Nineteen patients, all with progressive, heavily treated disease, were identified. Sites of tumor origin were: pancreas (74%), small bowel (11%), rectum (11%), and lung (5%); median Ki-67 was 32% (range 22-56). Thirteen patients (68%) completed all four 177Lu-DOTATATE cycles. Best response (N = 18 evaluable) was: 5/18 (28%) partial response, 8/18 (44%) stable disease, and 5/18 (28%) disease progression. Median PFS was 13.1 months (95% CI: 8.7-20.9). Most common treatment-related toxicities were thrombocytopenia (9 patients, 47%; G3/4, 1 patient, 5%), anemia (7 patients, 37%; G3/4, 2 patients, 11%), leukopenia (6 patients, 32%; G3/4, 0 patients), and liver function test elevation (4 patients, 21%; G3/4, 0 patients). NGS results were available from 13/19 tumors (68%). The most observed alterations were in MEN1 (6/13, 46%) and DAXX (4/13, 31%). No RB1 alterations identified.
We observed a meaningful disease control rate of 72% during treatment of WD HG NETs with 177Lu-DOTATATE. In this heavily pre-treated population, more than half of patients received all four treatment cycles with toxicities largely bone marrow-related. As would be expected in WD NETs, the vast majority had alterations in chromatin remodeling genes and no RB1 alterations.