Search

Cervical cancer is the fourth most common cancer in women worldwide. To determine early treatment for patients, it is critical to accurately classify the cervical intraepithelial lesion status based on a microscopic biopsy. Lesion classification is a 4-class problem, with biopsies being designated as benign or increasingly malignant as class 1-3, with 3 being invasive cancer. Unfortunately, traditional biopsy analysis by a pathologist is time-consuming and subject to intra- and inter-observer variability. For this reason, it is of interest to develop automatic analysis pipelines to classify lesion status directly from a digitalized whole slide image (WSI). The recent TissueNet Challenge was organized to find the best automatic detection pipeline for this task, using a dataset of 1015 annotated WSI slides. In this work, we present our winning end-to-end solution for cervical slide classification composed of a two-step classification model: First, we classify individual slide patches using an ensemble CNN, followed by an SVM-based slide classification using statistical features of the aggregated patch-level predictions. Importantly, we present the key innovation of our approach, which is a novel partial label-based loss function that allows us to supplement the supervised WSI patch annotations with weakly supervised patches based on the WSI class. This led to us not requiring additional expert tissue annotation, while still reaching the winning score of 94.7%. Our approach is a step towards the clinical inclusion of automatic pipelines for cervical cancer treatment planning.Clinical relevance- The explanation of the winning Tis-sueNet AI algorithm for automated cervical cancer classification, which may provide insights for the next generation of computer assisted tools in digital pathology.

The diagnosis of a uterine smooth muscle lesion is, in the majority of cases, straightforward. However, in a small number of cases, the morphological criteria used in such lesions cannot differentiate with certainty a benign from a malignant lesion and a diagnosis of smooth muscle tumor with uncertain malignant potential (STUMP) is made. Uterine leiomyosarcomas are often easy to diagnose but it is difficult or even impossible to identify a prognostic factor at the moment of the diagnosis with the exception of the stage. We hypothesize, for uterine smooth muscle lesions, that there is a gradient of genomic complexity that correlates to outcome. We first tested this hypothesis on STUMP lesions in a previous study and demonstrated that this 'gray category' could be split according to genomic index into two groups. A benign group, with a low to moderate alteration rate without recurrence and a malignant group, with a highly rearranged profile akin to uterine leiomyosarcomas. Here, we analyzed a large series of 77 uterine smooth muscle lesions (from 76 patients) morphologically classified as 19 leiomyomas, 14 STUMP and 44 leiomyosarcomas with clinicopathological and genomic correlations. We confirmed that genomic index with a cut-off=10 is a predictor of recurrence (P<0.0001) and with a cut-off=35 is a marker for poor overall survival (P=0.035). For the tumors confined to the uterus, stage as a prognostic factor was not useful in survival prediction. At stage I, among the tumors reclassified as molecular leiomyosarcomas (ie, genomic index ≥10), the poor prognostic markers were: 5p gain (overall survival P=0.0008), genomic index at cut-off=35 (overall survival P=0.0193), 13p loss including RB1 (overall survival P=0.0096) and 17p gain including MYOCD gain (overall survival P=0.0425). Based on these findings (and the feasibility of genomic profiling by array-comparative genomic hybridization), genomic index, 5p and 17p gains prognostic value could be evaluated in future prospective chemotherapy trials.

The diagnosis and management of uterine smooth muscle tumors with uncertain malignant potential (STUMP) is often challenging, and genomic data on these lesions as well as on uterine smooth muscle lesions are limited. We tested the hypothesis that genomic profile determination by array-CGH could split STUMP into a benign group with scarce chromosomal alterations akin to leiomyoma and a malignant group with high chromosomal instability akin to leiomyosarcoma. Array-CGH genomic profile analysis was conducted for a series of 29 cases of uterine STUMP. A group of ten uterine leiomyomas and ten uterine leiomyosarcomas served as controls. The mean age was 50 years (range, 24-85) and the follow-up ranged from 12 to 156 months (average 70 months). Since STUMP is a heterogenous group of tumors with genomic profiles that can harbor few to many chromosomal alterations, we compared genomic indices in leiomyomas and leiomyosarcomas and set a genomic index=10 threshold. Tumors with a genomic index <10 were classified as nonrecurring STUMPs and those with a genomic index >10 represented STUMPs with recurrences and unfavorable outcomes. Hence, the genomic index threshold splits the STUMP category into two groups of tumors with different outcomes: a group comparable to leiomyomas and another similar to leiomyosarcomas, but more indolent. In our STUMP series, genomic analysis by array-CGH is an innovative diagnostic tool for problematic smooth muscle uterine lesions, complementary to the morphological evaluation approach. We provide an improved classification method for distinguishing truly malignant tumors from benign lesions within the category of STUMP, especially those with equivocal morphological features.

The response of wheat crops to elevated CO 2 (eCO 2 ) was measured and modelled with the Australian Grains Free-Air CO 2 Enrichment experiment, located at Horsham, Australia. Treatments included CO 2 by water, N and temperature. The location represents a semi-arid environment with a seasonal VPD of around 0.5 kPa. Over 3 years, the observed mean biomass at anthesis and grain yield ranged from 4200 to 10 200 kg ha -1 and 1600 to 3900 kg ha -1 , respectively, over various sowing times and irrigation regimes. The mean observed response to daytime eCO 2 (from 365 to 550 μmol mol -1 CO 2 ) was relatively consistent for biomass at stem elongation and at anthesis and LAI at anthesis and grain yield with 21%, 23%, 21% and 26%, respectively. Seasonal water use was decreased from 320 to 301 mm (P = 0.10) by eCO 2 , increasing water use efficiency for biomass and yield, 36% and 31%, respectively. The performance of six models (APSIM-Wheat, APSIM-Nwheat, CAT-Wheat, CROPSYST, OLEARY-CONNOR and SALUS) in simulating crop responses to eCO 2 was similar and within or close to the experimental error for accumulated biomass, yield and water use response, despite some variations in early growth and LAI. The primary mechanism of biomass accumulation via radiation use efficiency (RUE) or transpiration efficiency (TE) was not critical to define the overall response to eCO 2 . However, under irrigation, the effect of late sowing on response to eCO 2 to biomass accumulation at DC65 was substantial in the observed data (~40%), but the simulated response was smaller, ranging from 17% to 28%. Simulated response from all six models under no water or nitrogen stress showed similar response to eCO 2 under irrigation, but the differences compared to the dryland treatment were small. Further experimental work on the interactive effects of eCO 2 , water and temperature is required to resolve these model discrepancies., (© 2014 John Wiley & Sons Ltd.)