Search

Interest for deep learning in radiology has increased tremendously in the past decade due to the high achievable performance for various computer vision tasks such as detection, segmentation, classification, monitoring, and prediction. This article provides step-by-step practical guidance for conducting a project that involves deep learning in radiology, from defining specifications, to deployment and scaling. Specifically, the objectives of this article are to provide an overview of clinical use cases of deep learning, describe the composition of multi-disciplinary team, and summarize current approaches to patient, data, model, and hardware selection. Key ideas will be illustrated by examples from a prototypical project on imaging of colorectal liver metastasis. This article illustrates the workflow for liver lesion detection, segmentation, classification, monitoring, and prediction of tumor recurrence and patient survival. Challenges are discussed, including ethical considerations, cohorting, data collection, anonymization, and availability of expert annotations. The practical guidance may be adapted to any project that requires automated medical image analysis.

Background: Patients with lesions in the posterosuperior (PS) segments of the liver have been considered poor candidates for laparoscopic liver resection (LLR). This study aims to compare short-term outcomes of LLR and open liver resections (OLR) in the PS segments., Methods: This multicenter study consisted of all patients who underwent LLR in the PS segments and all patients who underwent OLR in the PS segments between October 2011 and July 2016. Laparoscopic cases were case-matched with those who had an identical open procedure during the same period based on tumor location (same segment) and the Brisbane classification of the resection. Demographics, comorbid factors, perioperative outcomes, short-term outcomes, necessity of adjuvant chemotherapy, and the interval between surgery and initiation of adjuvant chemotherapy were compared between the two groups. Data were retrieved from a prospectively maintained electronic database., Results: Both groups were comparable for age, sex, ASA score, maximum tumor diameter, and number of patients with additional liver resections outside the posterior segments. Operative time was similar in both groups (median 140 min; p = 0.92). Blood loss was less in the LLR-group (median: 150 vs. 300 ml in OLR-group). Median hospital stay was 6 days in both groups. There was no significant difference in postoperative complications (OLR-group: 31.4% vs. LLR-group: 25.7%; p = 0.60). There was no significant difference in R0 resections (LLR: 97.2 vs. 100% in OLR; p = 1.00). Tumor-free margins were less in the LLR group (LLR: 5 vs. 9.5 mm in OLR; p = 0.012). Patients undergoing LLR were treated with chemotherapy sooner compared to those undergoing OLR (41 vs. 56 days, p = 0.02)., Conclusion: This study suggests that laparoscopic parenchymal preserving liver resections in the PS segments can be performed with comparable short-term outcomes as similar OLR. The shorter interval to chemotherapy might provide long-term oncologic benefits in patients who underwent LLR.

Objectives: Liver volumetry has emerged as an important tool in clinical practice. Liver volume is assessed primarily via organ segmentation of computed tomography (CT) and magnetic resonance imaging (MRI) images. The goal of this paper is to provide an accessible overview of liver segmentation targeted at radiologists and other healthcare professionals., Methods: Using images from CT and MRI, this paper reviews the indications for liver segmentation, technical approaches used in segmentation software and the developing roles of liver segmentation in clinical practice., Results: Liver segmentation for volumetric assessment is indicated prior to major hepatectomy, portal vein embolisation, associating liver partition and portal vein ligation for staged hepatectomy (ALPPS) and transplant. Segmentation software can be categorised according to amount of user input involved: manual, semi-automated and fully automated. Manual segmentation is considered the "gold standard" in clinical practice and research, but is tedious and time-consuming. Increasingly automated segmentation approaches are more robust, but may suffer from certain segmentation pitfalls. Emerging applications of segmentation include surgical planning and integration with MRI-based biomarkers., Conclusions: Liver segmentation has multiple clinical applications and is expanding in scope. Clinicians can employ semi-automated or fully automated segmentation options to more efficiently integrate volumetry into clinical practice., Teaching Points: • Liver volume is assessed via organ segmentation on CT and MRI examinations. • Liver segmentation is used for volume assessment prior to major hepatic procedures. • Segmentation approaches may be categorised according to the amount of user input involved. • Emerging applications include surgical planning and integration with MRI-based biomarkers.

Purpose: To compare the repeatability, agreement, and efficiency of MRI- and CT-based semiautomated liver segmentation for the assessment of total and subsegmental liver volume., Methods: This retrospective study was conducted in 31 subjects who underwent contemporaneous liver MRI and CT. Total and subsegmental liver volumes were segmented from contrast-enhanced 3D gradient-recalled echo MRI sequences and CT images. Semiautomated segmentation was based on variational interpolation and Laplacian mesh optimization. All segmentations were repeated after 2 weeks. Manual segmentation of CT images using an active contour tool was used as the reference standard. Repeatability and agreement of the methods were evaluated with intra-class correlation coefficients (ICC) and Bland-Altman analysis. Total interaction time was recorded., Results: Intra-reader ICC were ≥0.987 for MRI and ≥0.995 for CT. Intra-reader repeatability was 30 ± 217 ml (bias ± 1.96 SD) (95% limits of agreement: -187 to 247 ml) for MRI and -10 ± 143 ml (-153 to 133 ml) for CT. Inter-method ICC between semiautomated and manual volumetry were ≥0.995 for MRI and ≥0.986 for CT. Inter-method segmental ICC varied between 0.584 and 0.865 for MRI and between 0.596 and 0.890 for CT. Inter-method agreement was -14 ± 136 ml (-150 to 122 ml) for MRI and 50 ± 226 ml (-176 to 276 ml) for CT. Inter-method segmental agreement ranged from 10 ± 47 ml (-37 to 57 ml) to 2 ± 214 ml (-212 to 216 ml) for MRI and 9 ± 45 ml (-36 to 54 ml) to -46 ± 183 ml (-229 to 137 ml) for CT. Interaction time (mean ± SD) was significantly shorter for MRI-based semiautomated segmentation (7.2 ± 0.1 min, p < 0.001) and for CT-based semiautomated segmentation (6.5 ± 0.2 min, p < 0.001) than for CT-based manual segmentation (14.5 ± 0.4 min)., Conclusion: MRI-based semiautomated segmentation provides similar repeatability and agreement to CT-based segmentation for total liver volume.

Background: Liver volumetry is a critical component of safe hepatic surgery, in order to minimize the risk of postoperative liver failure. Liver volumes can be calculated routinely using the time-consuming gold standard method of manual volumetry. The current work sought to evaluate an alternative automatic technique based on a novel 3D virtual planning software, and to compare it to the manual technique., Methods: A prospective study of patients undergoing liver resection was conducted. Every patient had a pre and 2-day postoperative CT-scan. For each patient, total, remnant and resected volumes were calculated manually and automatically. Planes of resection were verified by a hepatobiliary surgeon and compared with postoperative volumes. Paired t-tests and correlation coefficients were calculated., Results: A major hepatectomy was carried out in 36/43 patients. The automatic TLV (1,759 mL) and the manual TLV (1,832 mL) were significantly different (p < 0.001), but extremely highly correlated (r = 0.989). The percentages of preoperative RLV (manual 58.5%, automatic 58.9%) were similar, with an excellent correlation of 0.917. The preoperative RLV were matched with the 2-day postoperative RLV showing a significant difference (p = 0.0301). The resected volumes using both techniques (871 and 832 mL) were compared with the resected specimen volume (670 mL), showing a significant difference (p < 0.001) but a high degree of correlation (r = 0.874)., Conclusion: The 3D virtual surgical planning software is accurate and reliable in determining the total liver and future remnant liver volumes. This technique demonstrates a good correlation with the manual technique. Future work will be required to confirm these findings and to evaluate the clinical value of the three-dimensional planning platform.

Diversified bet-hedging for life cycle duration is defined as within-generation variability in cycle length expressed by a single genotype maximising mean geometric fitness. Such plasticity is not predictive, i.e. it is not a response to cues from the environment that has a predictive value for the decision at hand. In evolutionary terms, diversified bet-hedging is perceived as an adaptation to environmental stochasticity. However, clear evidence of bet-hedging is scarce and exists only for a few desert plant species and one desert bee. In temperate insects, diversified bet-hedging for life cycle duration has been suspected in the chestnut weevil, but proximate factors responsible for individual variation are still unknown. From field experiments, we show that the frequency of the long cycle depends on larval weight and on the date when a larva abandons the fruit, but not on larval burying depth in the soil. Since the two first factors are known to depend on food and temperature and cannot lead to predictive plasticity, we give evidence of bet hedging in this temperate species. Indeed, despite a cost associated with prolonged diapause (extra mortality and loss of reproductive opportunity), a previous study showed that plasticity for life cycle duration, such as discussed in this paper, maximises mean geometric fitness and persistence probability in the chestnut weevil. We propose the hypothesis that the variation in life cycle duration depends on individual variability of metabolic resources such as lipids.

We test the adaptive value of clutch size observed in a natural population of the chestnut weevil Curculio elephas. Clutch size is defined as the number of immatures per infested chestnut. In natural conditions, clutch size averages 1.7 eggs. By manipulating clutch size in the field, we demonstrate that deviations from the theoretical "Lack clutch size", estimated as eight immatures, are mainly due to proximate and delayed effects of clutch size on offspring performance. We show the existence of a trade-off between clutch size and larval weight. The latter, a key life-history trait, is highly correlated with fitness because it is a strong determinant of larval survival and potential fecundity of offspring females. The fitness of different potential oviposition strategies characterized by their clutch sizes, ranging from one to nine immatures, was calculated from field- estimated parameters. Chestnut weevil females obtain an evolutionary advantage by laying their eggs singly, since, for instance, fitness of single-egg clutches exceeds fitness of two-egg clutches and four-egg clutches by 8.0% and 15.1% respectively.

Frequency distributions of insect immatures per host are often fitted to contagious distributions, such as the negative binomial, to deduce oviposition pattern. However, different mechanisms can be involved for each theoretical distribution and additional biological information is needed to correctly interpret the fits. We chose the chestnut weevil Curculio elephas, a pest of the European chestnut Castanea sativa, as a model to illustrate the difficulties of inferring oviposition pattern from fits to theoretical distributions and from the variance/mean ratio. From field studies over 13-16 years, we show that 20 out of the 31 yearly distributions available fit a negative binomial and 25 a zero-inflated Poisson (ZIP). No distribution fits a Poisson distribution. The ZIP distribution assumes heterogeneity within the fruit population. There are two categories of host: the first comprises chestnuts unsuitable for weevil oviposition or in excess relative to the number of weevil females, and the second comprises suitable fruits in which oviposition behavior is random. Our results confirm this host heterogeneity. According to the ZIP distribution, the first category of hosts includes on average 74% of the chestnuts. A negative binomial distribution may be generated by either true or false contagion. We show that neither interference between weevil females, nor spatial variation in the infestation rate exist. Consequently, the observed distributions of immatures are not the result of false contagion. Nevertheless, we cannot totally exlude true contagion of immatures. In this paper we discuss the difficulty of testing true contagion in natural conditions. These results show that we cannot systematically conclude in favour of contagion when fitting a distribution such as the negative binomial or when a variance/mean ratio is higher than unity.

Douglas fir seeds infested by diapausing larvae of a chalcid, Megastigmus spermotrophus, were transferred under different temperature and photoperiod regimes at various dates during autumn and winter. Both winter diapause and prolonged diapause were compared to the diapause patterns observed in control lots kept under natural conditions. The results showed that exposure to low temperatures in autumn and winter is required for the completion of diapause development. Median adult emergence dates and the date of transfer of larvae to 20°C were negatively correlated. The frequency of prolonged diapause of larvae transferred to 20°C decreased significantly after November. When larvae were submitted to different constant temperatures, the longer the daylength, the higher and faster the reactivation. These results suggest that chalcid larvae can perceive daylength through the seed coat. This phenomenon is probably related to the fact that most of the infested seeds falling in autumn are usually covered by a layer of leaf litter during the winter.

Detection of interspecific competition between insects is often sensitive to scaling. We give an example of scale-dependent interference between the weevil Curculio elephas and the moth Cydia splendana, which both have larvae that develop in the fruits of chestnut Castanea sativa. Measures at three scales were considered: chestnut, husk (with one to three fertile fruits) and tree. Data come from observations in the field over 14 years, complemented by experiments done directly in trees. Data on individual chestnut fruits revealed a marked statistical interference between the two insects. Experiments demonstrated that presence of a moth larva in a fruit usually inhibits weevil egg-laying. Conversely, weevil presence does not strongly modify moth larval behavior. Cases of double infestation often correspond to fruits first attacked by the weevil. With measures on husks, interference between the two insects was observed only in some trees; its intensity was always weaker than in the chestnuts themselves. At the scale of entire trees, rates of infestation by each insect are not correlated. Interference in chestnut fruits is interpreted by assuming that the weevil female either is sensitive to a repellent molecule originating from a moth larva or its frass, or can detect moth larval sounds. Mechanisms governing infestation rates from data per tree are discussed in relation to those found at fruit scale and to plant-insect interactions. The need to estimate available resources both from quantitative and qualitative points of view is emphasized.

In the chestnut weevil Curculio elephas, adult emergences spread over 3 or 4 years due to prolonged larval diapause in some individuals. Weevils with an extended diapause emerge, on the average, 1-10 days before those with simple diapause, but whatever the age of insects, emergences occur always from mid-August to early October. When the summer is dry, some adults cannot emerge because of the hardness of the soil. Emergence sucess of adults is smaller in females than in males. The result is that the sex ratio is female-biased before emergence and male-biased after. Summer drought cannot be predicted by the chestnut weevil, and when the soil is dry 27-78% of females cannot emerge and do not reproduce. The year after a summer drought, many reproducing females may emerge from larvae with prolonged diapause. These results suggest an evolutionary influence on the variability in diapause duration. Computer simulations and observations do not support the hypothesis that the main cause of variation in diapause length is the existence of several distinct genotypes within populations. On the contrary, our data strengthen the hypothesis for coin-flipping plasticity discussed in a previous paper.

Spreading of emergence over several years due to prolonged diapause in some larvae was shown in the chestnut weevil. Depending on the year the larvae buried themselves in the ground, 32-56% of live adults emerged after 2 or 3 years of underground life. Variability in the duration of diapause was assumed to correspond to tactics of adaptative "coin-flipping" plasticity. This plasticity must allow the chestnut weevil to respond to the unpredictability of its habitat as measured by the irregularity of chestnut production and summer drought. Indeed, fecundity and adult longevity did not lessen after 2 years of underground life. No drastic decrease in the population size of weevils occurs after bad years; for instance when the number of chestnuts on the study tree is less than 10 000, passers-by can collect all the fruit and about 95% of larvae developing in chestnuts are destroyed. Diapause nature (simple or prolonged) may be related to moisture and gas rates in the ground from October to December. These factors acting in autumn are not known to be involved in the physiological mechanisms that control the production of chestnuts.