Financial impact of incorporating deep learning reconstruction into magnetic resonance imaging routine

PURPOSE

Artificial intelligence and deep learning solutions are increasingly utilized in healthcare and radiology. The number of studies addressing their enhancement of productivity and monetary impact is, however, still limited. Our hospital has faced a need to enhance MRI scanner throughput, and we investigate the utility of new commercial deep learning reconstruction (DLR) algorithm for this purpose. In this work, a multidisciplinary team evaluated the impact of the widespread deployment of a new commercial deep learning reconstruction (DLR) algorithm for our magnetic resonance imaging scanner fleet.

METHODS

Our analysis centers on the DLR algorithm's effects on patient throughput and investment costs, contrasting these with alternative strategies for capacity expansion-namely, acquiring additional MRI scanners and increasing device utilization on weekends. We provide a framework for assessing the financial implications of new technologies in a trial phase, aiding in informed decision-making for healthcare investments.

RESULTS

We demonstrate substantial reductions in total operating costs compared to other capacity-enhancing methods. Specifically, the cost of adopting the deep learning technology for our entire scanner fleet is only 11 % compared to procuring an additional scanner and 20 % compared to the weekend utilization costs of existing devices.

CONCLUSIONS

Procuring DLR for our existing five-scanner fleet allows us to sustain our current MRI service levels without the need for an additional scanner, thereby achieving considerable cost savings. These reductions highlight the efficiency and economic viability of DLR in optimizing MRI service delivery.

Laparoscopic versus hybrid approach for treatment of incisional ventral hernia: a 5-10-year follow-up of the randomized controlled multicenter study

PURPOSE

In this long-term follow-up of a prospective, randomized, and multicenter study, we compare the results of a group receiving laparoscopic incisional ventral hernia repair using intraperitoneal onlay mesh (LG) to a group receiving a hybrid hernia repair where open closure of fascial defect was added to intraperitoneal mesh placement (HG).

METHODS

Originally, 193 patients with 2-7 cm incisional hernias were randomly assigned to either the LG or HG during the 30-month recruitment period in 2012 to 2015. Long-term follow-up was conducted 5-10 years after surgery to evaluate hernia recurrence rate and quality of life (QoL).

RESULTS

In all, 65 patients in the LG and 60 in the HG completed the long-term follow-up with a median follow-up period of 87 months. Recurrent hernia was detected in 11 of 65 patients (16.9%) in the LG and 10 of 60 patients (16.7%) in the HG (p  >  0.9). Kaplan-Meier analysis demonstrated a recurrence rate approaching 20% in both groups, with similar curves. Three patients in the LG (4.6% and five patients in the HG (8.1%) had undergone re-operation due to recurrence (p  = 0.48). There was no difference in patient-reported QoL measured using the SF-36 questionnaire. Mean pain scores were similar between groups, mean numeric rating scale (NRS) 0 to 10 being 1.1 in the LG and 0.7 in the HG (p  = 0.43).

CONCLUSION

Fascial closure did not reduce hernia recurrence rate in this study population, even though it has been shown to be beneficial and recommended in surgery guidelines. In the long term, recurrence rate for both groups is similar.

Deep-Learning-Based Contrast Synthesis From MRF Parameter Maps in the Knee Joint

BACKGROUND

Magnetic resonance fingerprinting (MRF) is a method to speed up acquisition of quantitative MRI data. However, MRF does not usually produce contrast-weighted images that are required by radiologists, limiting reachable total scan time improvement. Contrast synthesis from MRF could significantly decrease the imaging time.

PURPOSE

To improve clinical utility of MRF by synthesizing contrast-weighted MR images from the quantitative data provided by MRF, using U-nets that were trained for the synthesis task utilizing L1- and perceptual loss functions, and their combinations.

STUDY TYPE

Retrospective.

POPULATION

Knee joint MRI data from 184 subjects from Northern Finland 1986 Birth Cohort (ages 33-35, gender distribution not available).

FIELD STRENGTH AND SEQUENCE

A 3 T, multislice-MRF, proton density (PD)-weighted 3D-SPACE (sampling perfection with application optimized contrasts using different flip angle evolution), fat-saturated T2-weighted 3D-space, water-excited double echo steady state (DESS).

ASSESSMENT

Data were divided into training, validation, test, and radiologist's assessment sets in the following way: 136 subjects to training, 3 for validation, 3 for testing, and 42 for radiologist's assessment. The synthetic and target images were evaluated using 5-point Likert scale by two musculoskeletal radiologists blinded and with quantitative error metrics.

STATISTICAL TESTS

Friedman's test accompanied with post hoc Wilcoxon signed-rank test and intraclass correlation coefficient. The statistical cutoff P <0.05 adjusted by Bonferroni correction as necessary was utilized.

RESULTS

The networks trained in the study could synthesize conventional images with high image quality (Likert scores 3-4 on a 5-point scale). Qualitatively, the best synthetic images were produced with combination of L1- and perceptual loss functions and perceptual loss alone, while L1-loss alone led to significantly poorer image quality (Likert scores below 3). The interreader and intrareader agreement were high (0.80 and 0.92, respectively) and significant. However, quantitative image quality metrics indicated best performance for the pure L1-loss.

DATA CONCLUSION

Synthesizing high-quality contrast-weighted images from MRF data using deep learning is feasible. However, more studies are needed to validate the diagnostic accuracy of these synthetic images.

EVIDENCE LEVEL

4.

TECHNICAL EFFICACY

Stage 1.

Critical evaluation of deep neural networks for wrist fracture detection

Wrist Fracture is the most common type of fracture with a high incidence rate. Conventional radiography (i.e. X-ray imaging) is used for wrist fracture detection routinely, but occasionally fracture delineation poses issues and an additional confirmation by computed tomography (CT) is needed for diagnosis. Recent advances in the field of Deep Learning (DL), a subfield of Artificial Intelligence (AI), have shown that wrist fracture detection can be automated using Convolutional Neural Networks. However, previous studies did not pay close attention to the difficult cases which can only be confirmed via CT imaging. In this study, we have developed and analyzed a state-of-the-art DL-based pipeline for wrist (distal radius) fracture detection-DeepWrist, and evaluated it against one general population test set, and one challenging test set comprising only cases requiring confirmation by CT. Our results reveal that a typical state-of-the-art approach, such as DeepWrist, while having a near-perfect performance on the general independent test set, has a substantially lower performance on the challenging test set-average precision of 0.99 (0.99-0.99) versus 0.64 (0.46-0.83), respectively. Similarly, the area under the ROC curve was of 0.99 (0.98-0.99) versus 0.84 (0.72-0.93), respectively. Our findings highlight the importance of a meticulous analysis of DL-based models before clinical use, and unearth the need for more challenging settings for testing medical AI systems.

Discrepancies in interpretation of night-time emergency computed tomography scans by radiology residents

Background

In many emergency radiology units, most of the night-time work is performed by radiology residents. Residents’ preliminary reports are typically reviewed by an attending radiologist. Accordingly, it is known that discrepancies in these preliminary reports exist.

Purpose

To evaluate the quality of night-time computed tomography (CT) interpretations made by radiology residents in the emergency department.

Material and Methods

Retrospectively, 1463 initial night-time CT interpretations given by a radiology resident were compared to the subspecialist’s re-interpretation given the following weekday. All discrepancies were recorded and classified into different groups regarding their possible adverse effect for the emergency treatment. The rate of discrepancies was compared between more and less experienced residents and between different anatomical regions.

Results

The overall rate of misinterpretations was low. In 2.3% (33/1463) of all night-time CT interpretations, an important and clinically relevant diagnosis was missed. No fatalities occurred due to CT misinterpretations during the study. The total rate of discrepancies including clinically irrelevant findings such as anatomical variations was 12.2% (179/1463). Less experienced residents were more likely to miss the correct diagnosis than more experienced residents (18.3% vs. 10.9%, odds ratio [OR] = 1.82, P = 0.001). Discrepancies were more common in body CT interpretations than in neurological CTs (18.1% vs. 9.1%, OR = 2.30, P < 0.0001).

Conclusion

The rate of clinically important misinterpretations in CT examinations by radiology residents was found to be low. Experience helps in lowering the rate of misinterpretations.

Direct interaction of actin filaments with F-BAR protein pacsin2

Two mechanisms have emerged as major regulators of membrane shape: BAR domain-containing proteins, which induce invaginations and protrusions, and nuclear promoting factors, which cause generation of branched actin filaments that exert mechanical forces on membranes. While a large body of information exists on interactions of BAR proteins with membranes and regulatory proteins of the cytoskeleton, little is known about connections between these two processes. Here, we show that the F-BAR domain protein pacsin2 is able to associate with actin filaments using the same concave surface employed to bind to membranes, while some other tested N-BAR and F-BAR proteins (endophilin, CIP4 and FCHO2) do not associate with actin. This finding reveals a new level of complexity in membrane remodeling processes.

Crystallization and phasing of focal adhesion protein 52 fromGallus gallus

Focal adhesion protein 52 (FAP52) is a multidomain adaptor protein of 448 amino acids characterized as an abundant component of focal adhesions. FAP52 binds to filamin via its N-terminal alpha-helical domain, suggesting a role in linking focal adhesions to the actin-based cytoskeleton. The recombinant protein was crystallized using the hanging-drop vapour-diffusion method, which yielded two crystal forms. Native data were collected from both crystal forms to 2.8 and 2.1 A resolution, respectively. For one of the crystal forms, initial MAD phasing was successfully performed using two data sets from xenon-derivatized crystals. The derivative data sets were collected using softer X-rays of 1.5 and 1.9 A wavelength. Preliminary structural analysis reveals the presence of a dimer in the asymmetric unit.

Focal adhesion protein FAP52 self-associates through a sequence conserved among the members of the PCH family proteins

FAP52 is a recently described focal adhesion-associated protein. It is a member of an emerging PCH (pombe Cdc15 homology) family of proteins characterized by a common domain organization and involvement in actin cytoskeleton organization, cytokinesis, and vesicular trafficking. Using gel filtration, surface plasmon resonance, and native polyacrylamide gel electrophoresis analysis, combined with chemical cross-linking of both native and recombinant protein, we show that FAP52 self-associates in vitro and suggest that it occurs predominantly as a trimer also in vivo. Analysis of the various domains of FAP52 by surface plasmon resonance showed that the highly alpha-helical region in the N-terminal half of the protein provides the self-association interface. Overexpression of the oligomerization domain in cultured cells was accompanied by major alterations in cellular morphology, actin organization, and the structure of focal adhesions, suggesting that an orderly coming together of FAP52 molecules is crucial for a proper actin filament organization and cytoskeletal structure. Comparison of the primary structures shows that all of the members of the PCH family have, in their N-terminal halves, a similar, highly alpha-helical region, suggesting that they all have a capacity to self-associate.

FAP52 regulates actin organization via binding to filamin

FAP52, a focal adhesion-associated phosphoprotein, is a member of a FAP52/PACSIN/syndapin family of proteins. They share a multidomain structure and are implicated in actin-based and endocytotic functions. We show, by using both native and recombinant proteins, that FAP52 selectively binds to the actin cross-linking protein filamin (ABP-280). This was based on an affinity purification followed by a sequence determination by mass spectrometry, co-immunoprecipitation, overlay binding, and surface plasmon resonance analysis. Binding studies with deletion mutants showed that the sites of the interaction map to the highly alpha-helical N-terminal part of FAP52 and to the C-terminal region of filamin, which also contains binding sites to some transmembrane signaling proteins. In immunofluorescence and immunoelectron microscopy of cultured fibroblasts, a different overall subcellular distribution was seen for filamin and FAP52 except for a stress fiber-focal adhesion junction where they showed a notable overlap. Overexpression of the full-length and mutant forms of FAP52 led to an extensive reorganization of actin and filamin in cultured fibroblasts. Thus, the results show that FAP52 interacts with filamin, and we propose that this interaction is important in linking and coordinating the events between focal adhesions and the actin cytoskeleton.