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.

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 Ͻ2 Å RMSD error for the mainchain N, C a , C, and O atoms; the average accuracies were 0.59 6 0.05, 1.16 6 0.10, and 2.61 6 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.

Background FOLFIRINOX therapy for pancreatic ductal adenocarcinoma (PDAC) has been reported to result in objective response rates that are 2–3 fold higher than other regimens. Our goal was to assess response rates and resection rates in locally unresectable (stage III) patients initially treated with induction FOLFIRINOX. Methods The institutional cancer database was queried for patients treated with induction FOLFIRINOX therapy between 2010–2013. Patients were included if they were treated at our institution for stage III PDAC (locally unresectable) that had been adjudicated at a weekly multidisciplinary tumor board. Results One hundred and one patients were identified. The median age was 64 years (range:37–81) and the median follow-up was 12 months (range:3–37). Patients received a median of six cycles of induction FOLFIRINOX(range:1–20). No grade 4–5 toxicity was recorded. At initial restaging(median of 3 months following diagnosis), 23 patients(23%) had developed distant metastases, 15 patients(15%) underwent resection, and 63 patients(63%) proceeded to chemoradiation. Within the group of 63patients who proceeded to chemoradiation(median of 9 months following diagnosis), an additional 16 patients(16%) underwent resection and 5(5%) developed metastases. A partial radiographic response was observed in 29% of all patients, which was associated with the ability to perform resection(p=0.004). The median overall survival within the group who progressed on FOLFIRINOX and the group who did not progress were 11 and 26months, respectively. Conclusion Nearly a third of patients who had been initially identified to have stage III pancreatic carcinoma and had been treated with FOLFIRINOX responded radiographically and underwent tumor resection.

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.

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).