Search

We introduce an end-to-end unsupervised (or weakly supervised) image registration method that blends conventional medical image registration with contemporary deep learning techniques from computer vision. Our method downsamples both the fixed and the moving images into multiple feature map levels where a displacement field is estimated at each level and then further refined throughout the network. We train and test our model on three different datasets. In comparison with the initial registrations we find an improved performance using our model, yet we expect it would improve further if the model was fine-tuned for each task. The implementation is publicly available (https://github.com/ngunnar/learning-a-deformable-registration-pyramid).

Background: In a recent, large population-based study we identified an increased risk of a second malignancy in patients following a diagnosis of Chronic Myeloid Leukemia (CML), as compared to an age- and gender-matched control cohort (Gunnarsson et al, Br J Haematol. 2015 Jun; 169(5): 683-8). If CML patients have an increased congenital or acquired susceptibility to develop any cancer, the prevalence of other prior malignancies would be expected to be increased already at the time of CML diagnosis. There is a known association between autoimmune disease (AD) or chronic inflammatory disease (CID) and the development of some hematological malignancies. However, to date, studies on the prevalence of AD or CID detected prior to the CML diagnosis are few and inconclusive. Our aim was to estimate the prevalence of other malignancies, AD or CID in CML patients at or before the time point of the CML diagnosis. Materials and methods: We used the population-based Swedish CML Register to identify patients diagnosed with CML in Sweden between 2002-2013. This cohort was linked to the Swedish Cancer Register to retrieve information about malignancies reported before the diagnosis of CML and the Swedish National Patient Register to retrieve information about AD and CID. For each of the 984 patients with CML, five age-, gender- and county of residence-matched controls were selected from the general population. All controls had to be free of CML and alive at the time of CML diagnosis for the corresponding case patient. Conditional logistic regression was used to calculate odds ratios (OR) and 95% confidence intervals (CI). We excluded diagnoses registered during the year prior to the CML diagnosis to avoid detection bias. In separate steps, analyses were also performed based on diagnoses of AD, CID or cancer three years before the date of CML diagnosis. Results: A total of 984 CML patients were assessed with regard to a prior diagnosis of malignancy, AD or CID excluding the year immediately prior to their CML diagnosis, representing more than 45.000 person-years of follow-up. Compared to the matched population controls, the prevalence of prior malignancies and AD were elevated in CML patients: OR 1.47 (95%CI 1.20-1.82) and 1.55 (1.21-1.98), respectively. Breast-, gastrointestinal- and urinary tract cancers and melanomas were common cancer types and were all significantly more prevalent in the CML cohort, table I. After implementing a three-year exclusion period before the date of CML diagnosis, prior malignancies remained more prevalent in CML patients. Assessment of ADs was hampered by small numbers, sarcoidosis was the only AD with increased prevalence: OR 13.43; 95 % CI 3.56 - 50.73. No association was detected between CML and previous CID. Conclusions: Based on a large population-based cohort, our findings indicate that CML patients have an increased prevalence of other malignancies and AD prior to the diagnosis of CML, suggesting that a hereditary or acquired predisposition to cancer and/or autoimmunity is involved in the pathogenesis of CML. Table 1. Odds Ratios for malignancies prior to CML among 984 Swedish CML patients diagnosed between 2002 and 2013. Diagnoses of malignancy during the year immediately prior to CML diagnosis excluded to avoid detection bias. Participants Latency More Than 3 YearsBefore CML Diagnosis Variable CMLn=984 Controlsn = 4920 OR 95% CI CMLn=984 Controlsn = 4920 OR 95% CI Overall 128 453 1.47 1.20 - 1.82 113 381 1.55 1.24 - 1.93 Men 58 184 1.61 1.19 - 2.18 50 146 1.75 1.26 - 2.43 Women 70 269 1.32 1.01 - 1.74 63 235 1.36 1.02 - 1.82 Age Disclosures Richter: Ariad: Honoraria; Bristol-Myers Squibb: Honoraria; Novartis: Honoraria. Sjalander:Novartis: Honoraria.

The clinical outcome for patients with chronic myeloid leukemia (CML) has improved dramatically following the introduction of tyrosine kinase inhibitors. An improved survival, combined with a constant incidence, is expected to increase the prevalence of CML. However, data on the prevalence of CML remain scarce. We examined the overall and relative (age and gender matched) survival and assessed the past, present, and projected future prevalence of CML in Sweden. Data on all patients diagnosed with CML between 1970 and 2012 were retrieved from the Swedish Cancer Register and the Swedish Cause of Death Register. The 5-year overall survival increased from 0.18 to 0.82, during the observed time period. Between 2006 and 2012, the 5-year relative survival was close to normal for 40-year-old, but considerably lower for 80-year-old CML patients. The observed prevalence tripled from 1985 to 2012, from 3.9 to 11.9 per 100 000 inhabitants. Assuming no further improvements in relative survival, the prevalence is projected to further increase by 2060 to 22.0 per 100 000 inhabitants (2587 persons in Sweden). The projected dramatic increase in CML prevalence has major medical and health economic implications and needs to be considered in planning how to organize future care of CML patients.

No increased prevalence of malignancies among first-degree relatives of 800 patients with chronic myeloid leukemia : a population-based study in Sweden

Background: The etiology of chronic myeloid leukemia (CML) is essentially unknown with high doses of ionizing radiation being the only well established risk factor. We have recently published two large population-based studies showing an increased prevalence of other malignancies prior, as well as subsequent to a diagnosis of CML (Gunnarsson et al, Br J Haematol. 2015 Jun; 169(5): 683-8 and Gunnarsson et al, Leukemia. 2016 Jul; 30(7): 1562-7). One may therefore speculate that CML patients may have an increased congenital or acquired susceptibility to develop cancer. In the former case, one would expect an increased prevalence of malignancies among first-degree relatives (FDR) to CML patients. In a previous report based on the Swedish Cancer Registry, no increased aggregation of malignancies was detected among family members to CML patients diagnosed between 1958 and 2004 (Bjorkholm et al, Blood. 2013 Jul 18; 122(3): 460-1). However, a more strict definition of CML (requiring e.g. thepresence of a Philadelphia chromosome or the BCR/ABL fusion gene) was introduced with the updated WHO classification in 2002, making subsequently diagnosed CML cohorts more homogenous. Materials and methods: Aiming to establish the prevalence of malignancies among FDR of a large and well-defined contemporary CML cohort in Sweden compared to carefully matched controls, we have used four large Swedish population based registers. To identify Swedish patients with CML diagnosed between 2002 and 2013, we used the Swedish CML Register to which all CML patients diagnosed January 1st 2002 and later are reported. FDR were identified by use of the Swedish Multi-Generation Register, which comprises information about parent-sibling-offspring relationships of persons that has been registered in Sweden at some time since 1961 and born later than 1932. By linking this cohort to the Swedish Cancer Register, we retrieved information about malignancies for each FDR. Each CML patient was matched with five, age-, gender- and county of residence-matched controls, selected from the Swedish Total Population Register. All controls had to be alive and free of CML at the time of CML diagnosis for the matching CML patient. To calculate odds ratio (OR) and 95% confidence intervals (CI), conditional logistic regression were used. Results: In the Swedish CML register, 984 patients were identified. Among them 184 patients were excluded due to a birth year prior to 1932. Among the 800 remaining CML patients, 4 287 FDR were identified and included in the analysis (parents: 1 346, siblings: 1 497 and children: 1 444). Correspondingly, 20 930 matched controls were included in the analysis. In total, 611 malignancies were identified among the FDR of CML patients compared to 2844 in the control group yielding an OR of 1.057 (95% CI 0.962 - 1.162). Neither hematological malignancies nor solid tumors were increased in the CML-FDR group (Table 1). Notably, none of the FDRs in the CML-FDR group had a CML diagnosis. Conclusions: Using data from four large Swedish population based registers and based on the fate of more than 4 000 FDR of 800 CML patients diagnosed in the modern era of cytogenetics, as well as closely matched CML-free controls, we show that there is no familial aggregation of malignancies in FDRs of patients with CML. These results suggest that a hereditary predisposition to develop cancer is unlikely to be a part of the pathogenesis of CML. Disclosures Höglund: Akinion Pharmaceuticals: Consultancy; Janssen-Cilag: Honoraria. Lambe:AstraZeneca: Other: Stock Ownership ; Pfizer: Other: Stock Ownership . Richter:Pfizer: Honoraria, Research Funding; BMS: Honoraria, Research Funding; Novartis: Honoraria, Research Funding; Ariad: Honoraria, Research Funding. Själander:ARIAD: Consultancy.