Dynamic contrast-enhanced computed tomography diagnosis of primary liver cancers using transfer learning of pretrained convolutional neural networks: Is registration of multiphasic images necessary?

PURPOSE: To evaluate the effect of image registration on the diagnostic performance of transfer learning (TL) using pretrained convolutional neural networks (CNNs) and three-phasic dynamic contrast-enhanced computed tomography (DCE-CT) for primary liver cancers. METHODS: We retrospectively evaluated 215 consecutive patients with histologically proven primary liver cancers, including six early, 58 well-differentiated, 109 moderately differentiated, 29 poorly differentiated hepatocellular carcinomas (HCCs), and 13 non-HCC malignant lesions containing cholangiocellular components. We performed TL using various pretrained CNNs and preoperative three-phasic DCE-CT images. Three-phasic DCE-CT images were manually registered to correct respiratory motion. The registered DCE-CT images were then assigned to the three color channels of an input image for TL: pre-contrast, early phase, and delayed phase images for the blue, red, and green channels, respectively. To evaluate the effects of image registration, the registered input image was intentionally misaligned in the three color channels by pixel shifts, rotations, and skews with various degrees. The diagnostic performances (DP) of the pretrained CNNs after TL in the test set were compared by three general radiologists (GRs) and two experienced abdominal radiologists (ARs). The effects of misalignment in the input image and the type of pretrained CNN on the DP were statistically evaluated. RESULTS: The mean DPs for histological subtype classification and differentiation in primary malignant liver tumors on DCE-CT for GR and AR were 39.1%, and 47.9%, respectively. The highest mean DPs for CNNs after TL with pixel shifts, rotations, and skew misalignments were 44.1%, 44.2%, and 43.7%, respectively. Two-way analysis of variance revealed that the DP is significantly affected by the type of pretrained CNN (P = 0.0001), but not by misalignments in input images other than skew deformations. CONCLUSION: TL using pretrained CNNs is robust against misregistration of multiphasic images and comparable to experienced ARs in classifying primary liver cancers using three-phasic DCE-CT.

Usefulness of three-dimensional magnetic resonance cholangiopancreatography with partial maximum intensity projection for diagnosing autoimmune pancreatitis.

PURPOSE: To compare three-dimensional magnetic resonance cholangiopancreatography (MRCP) with/without partial maximum intensity projection (MIP) and endoscopic retrograde cholangiopancreatography (ERCP) in patients with autoimmune pancreatitis (AIP). MATERIALS AND METHODS: Three-dimensional MRCP and ERCP images were retrospectively analyzed in 24 patients with AIP. We evaluated the narrowing length of the main pancreatic duct (NR-MPD), multiple skipped MPD narrowing (SK-MPD), and side branches arising from the narrowed portion of the MPD (SB-MPD) using four MRCP datasets: 5 original images (MIP5), 10 original images (MIP10), all original images (full-MIP), and a combination of these three datasets (a-MIP). The images were scored using a 3- or 5-point scale. The scores of the four MRCP datasets were statistically analyzed, and the positive rate of each finding was compared between MRCP and ERCP. RESULTS: The median scores for SB-MPD on MIP5 and a-MIP were significantly higher than those on MIP10 and full-MIP. In other words, partial MIP is superior to full-MIP for visualization of detailed structures. The positive rate for SB-MPD on full-MIP was significantly lower than that on ERCP, whereas the positive rate on MIP5, MIP10, and a-MIP was not significantly different from that on ERCP. Moreover, the positive rate for NR-MPD and SK-MPD on the MRCP images was significantly higher than that on the ERCP images. CONCLUSION: Partial MIP is useful for evaluating the MPD and is comparable with ERCP for diagnosing AIP.

Constructing biological pathway models with hybrid functional petri nets.

In many research projects on modeling and analyzing biological pathways, the Petri net has been recognized as a promising method for representing biological pathways. From the pioneering works by Reddy et al., 1993, and Hofestädt, 1994, that model metabolic pathways by traditional Petri net, several enhanced Petri nets such as colored Petri net, stochastic Petri net, and hybrid Petri net have been used for modeling biological phenomena. Recently, Matsuno et al., 2003b, introduced the hybrid functional Petri net (HFPN) in order to give a more intuitive and natural modeling method for biological pathways than these existing Petri nets. Although the paper demonstrates the effectiveness of HFPN with two examples of gene regulation mechanism for circadian rhythms and apoptosis signaling pathway, there has been no detailed explanation about the method of HFPN construction for these examples. The purpose of this paper is to describe method to construct biological pathways with the HFPN step-by-step. The method is demonstrated by the well-known glycolytic pathway controlled by the lac operon gene regulatory mechanism.

Synthesis and biological activity of novel alpha-substituted beta-phenylpropionic acids having pyridin-2-ylphenyl moiety as antihyperglycemic agents.

We previously reported the identification of novel oximes having 5-benzyl-2,4-thiazolidinedione with antihyperglycemic activity. We now report the synthesis and biological activity of a novel series of oximes and amides having alpha-substituted-beta-phenylpropionic acids. In this series, we obtained potent PPAR alpha/gamma dual agonist (S)-9d, with which activation of PPAR alpha and PPAR gamma was considerably more potent than that of the reference compounds GW9578 22 and rosiglitazone 3, respectively. This means (S)-9d is of the strongest class of PPAR alpha/gamma dual agonists. In the course of this study, we also obtained 8h, which indicated potent plasma glucose lowering effect in spite of weak PPAR alpha/gamma agonistic activity.

Constructing biological pathway models with hybrid functional Petri nets.

In many research projects on modeling and analyzing biological pathways, the Petri net has been recognized as a promising method for representing biological pathways. From the pioneering works by Reddy et al., 1993, and Hofestädt, 1994, that model metabolic pathways by traditional Petri net, several enhanced Petri nets such as colored Petri net, stochastic Petri net, and hybrid Petri net have been used for modeling biological phenomena. Recently, Matsuno et al., 2003b, introduced the hybrid functional Petri net (HFPN) in order to give a more intuitive and natural modeling method for biological pathways than these existing Petri nets. Although the paper demonstrates the effectiveness of HFPN with two examples of gene regulation mechanism for circadian rhythms and apoptosis signaling pathway, there has been no detailed explanation about the method of HFPN construction for these examples. The purpose of this paper is to describe method to construct biological pathways with the HFPN step-by-step. The method is demonstrated by the well-known glycolytic pathway controlled by the lac operon gene regulatory mechanism.

Boundary formation by notch signaling in Drosophila multicellular systems: experimental observations and gene network modeling by Genomic Object Net.

The Delta-Notch signaling system plays an essential role in various morphogenetic systems of multicellular animal development. Here we analyzed the mechanism of Notch-dependent boundary formation in the Drosophila large intestine, by experimental manipulation of Delta expression and computational modeling and simulation by Genomic Object Net. Boundary formation representing the situation in normal large intestine was shown by the simulation. By manipulating Delta expression in the large intestine, a few types of disorder in boundary cell differentiation were observed, and similar abnormal patterns were generated by the simulation. Simulation results suggest that parameter values representing the strength of cell-autonomous suppression of Notch signaling by Delta are essential for generating two different modes of patterning: lateral inhibition and boundary formation, which could explain how a common gene regulatory network results in two different patterning modes in vivo. Genomic Object Net proved to be a useful and flexible biosimulation system that is suitable for analyzing complex biological phenomena such as patternings of multicellular systems as well as intracellular changes in cell states including metabolic activities, gene regulation, and enzyme reactions.

Genomic Object Net: II. Modelling biopathways by hybrid functional Petri net with extension.

This paper demonstrates how to create an HFPNe (hybrid functional Petri net with extension) model, using the lac operon gene regulatory mechanism and glycolytic pathway as an example. Using this example, readers can then model other biopathways of interest. Simulations of the HFPNe model were performed using the software package Genomic Object Net.