Doing More With Less: A Multitask Deep Learning Approach in Plant Phenotyping.

Image-based plant phenotyping has been steadily growing and this has steeply increased the need for more efficient image analysis techniques capable of evaluating multiple plant traits. Deep learning has shown its potential in a multitude of visual tasks in plant phenotyping, such as segmentation and counting. Here, we show how different phenotyping traits can be extracted simultaneously from plant images, using multitask learning (MTL). MTL leverages information contained in the training images of related tasks to improve overall generalization and learns models with fewer labels. We present a multitask deep learning framework for plant phenotyping, able to infer three traits simultaneously: (i) leaf count, (ii) projected leaf area (PLA), and (iii) genotype classification. We adopted a modified pretrained ResNet50 as a feature extractor, trained end-to-end to predict multiple traits. We also leverage MTL to show that through learning from more easily obtainable annotations (such as PLA and genotype) we can predict a better leaf count (harder to obtain annotation). We evaluate our findings on several publicly available datasets of top-view images of Arabidopsis thaliana. Experimental results show that the proposed MTL method improves the leaf count mean squared error (MSE) by more than 40%, compared to a single task network on the same dataset. We also show that our MTL framework can be trained with up to 75% fewer leaf count annotations without significantly impacting performance, whereas a single task model shows a steady decline when fewer annotations are available. Code available at https://github.com/andobrescu/Multi_task_plant_phenotyping.

A novel quantitative method for estimating bone mineral density using B-mode ultrasound and radiofrequency signals-a pilot study on patients with rheumatoid arthritis.

The objective of the present study was to evaluate whether an innovative quantitative ultrasound (QUS) technique, Radiofrequency Echographic Multi Spectromety, which combines B-mode ultrasound and radiofrequency signals, is reliable in typical Romanian patients compared to previous results obtained using dual-energy X-ray absorptiometry (DXA). The study prospectively included previously unscreened post-menopausal females with rheumatoid arthritis (RA) and age-matched healthy controls. Bone mineral density (BMD) measurements were performed with an EchoS machine (Echolight®), which combines B-mode ultrasound and radiofrequency signals. The study included 106 RA patients, with a median disease duration of 3.2 (0.5-22) years and 119 controls. RA patients had a significantly lower body weight, body mass index (BMI) and basal metabolic rate (BMR) than the controls, while the prevalence of obesity and body fat differed insignificantly. RA patients had a significantly lower spine and hip BMD, higher fracture risk and higher prevalence of osteoporosis. Compared to RA patients without osteoporosis, those with osteoporosis were significantly older and had a longer menopause duration, but they had a significantly lower BMI, body fat, BMR and prevalence of obesity. Among the controls and RA patients, the median spine and hip BMD became significantly higher as the BMI increased from underweight to obesity. In conclusion, osteoporosis is prevalent among RA patients, as a part of a complex transformation of body mass composition, involving BMI and fat mass. The novel QUS scanning technique was able to replicate the results of the established DXA measurement of BMD and is potentially suitable for screening wide populations for osteoporosis.

Normal and pathological aging of attention in presymptomatic Huntington's, Huntington's and Alzheimer's Disease, and nondemented elderly subjects.

BACKGROUND: Attention models view attention as having at least two components: endogenous attention defined as executive and directed by voluntary acts, and exogenous attention defined as automatic and directed by external stimulation. METHODS: Three studies (2 of our own) were designed to evaluate the decline of these two components of attention in normal aging and two neurodegenerative diseases. Standardized tests derived from Posner's model of visuospatial attention were administered to normal healthy elderly participants (n = 13), patients suffering from Huntington's disease (HD; n = 17) and Alzheimer's disease (n = 15), and matched control subjects (n = 57). Outcome measures were reaction time (RT) and RT difference score (defined as invalid RT minus valid RT). RESULTS: In healthy elderly participants, the decline was more pronounced for endogenous attention in situations of perceptual conflict. In Alzheimer's disease, there was a significant decline in both attention components, while in HD, voluntary attention was markedly impaired and automatic attention preserved. CONCLUSIONS: Normal aging and HD are characterized by decreased endogenous attention in situations of perceptual conflict. Our data support previous findings that older people display impairment of attention in complex perceptual situations. We propose a model which allows for the separation of attention pathologies, thus improving therapeutic strategies for patients and elderly.

A "Do-It-Yourself" phenotyping system: measuring growth and morphology throughout the diel cycle in rosette shaped plants.

BACKGROUND: Improvements in high-throughput phenotyping technologies are rapidly expanding the scope and capacity of plant biology studies to measure growth traits. Nevertheless, the costs of commercial phenotyping equipment and infrastructure remain prohibitively expensive for wide-scale uptake, while academic solutions can require significant local expertise. Here we present a low-cost methodology for plant biologists to build their own phenotyping system for quantifying growth rates and phenotypic characteristics of Arabidopsis thaliana rosettes throughout the diel cycle. RESULTS: We constructed an image capture system consisting of a near infra-red (NIR, 940 nm) LED panel with a mounted Raspberry Pi NoIR camera and developed a MatLab-based software module (iDIEL Plant) to characterise rosette expansion. Our software was able to accurately segment and characterise multiple rosettes within an image, regardless of plant arrangement or genotype, and batch process image sets. To further validate our system, wild-type Arabidopsis plants (Col-0) and two mutant lines with reduced Rubisco contents, pale leaves and slow growth phenotypes (1a3b and 1a2b) were grown on a single plant tray. Plants were imaged from 9 to 24 days after germination every 20 min throughout the 24 h light-dark growth cycle (i.e. the diel cycle). The resulting dataset provided a dynamic and uninterrupted characterisation of differences in rosette growth and expansion rates over time for the three lines tested. CONCLUSION: Our methodology offers a straightforward solution for setting up automated, scalable and low-cost phenotyping facilities in a wide range of lab environments that could greatly increase the processing power and scalability of Arabidopsis soil growth experiments.

Coronary Calcium Scanning in Patients after Adjuvant Radiation for Early Breast Cancer and Ductal Carcinoma In situ.

BACKGROUND AND OBJECTIVE: Radiation therapy (RT) is part of standard adjuvant treatment for breast cancer. Earlier studies demonstrated increased cardiac morbidity and mortality from this. Coronary Calcium scanning utilizing Multidetector Computed Tomography (MDCT) can detect early atherosclerosis in coronary arteries by identifying the amount of calcifications. In our study we employed these tools to detect occult atherosclerosis at least 5 years following breast RT. METHODS: We evaluated 20 asymptomatic patients, <60 years old, treated with RT at least 5 years prior to enrollment. Nine received RT to the left and 11 to the right chest wall. The median interval between RT and calcium scan was 8 years. All patients were treated with external beam RT using tangential technique. All patients underwent MDCT to compute volumetric and Agatston calcium scores of the coronary arteries and the aorta. RESULTS: Eleven patients had RT to the right chest wall, and eight had a calcium score of 0, while two had minimally elevated scores and one patient had a significantly elevated score. Meanwhile nine patients had RT to the left chest wall, and seven had a calcium score of 0. None had significantly elevated scores. In the aorta, 11 of 20 patients had a score of 0, while 8 of 20 had minimally elevated scores. CONCLUSION: In contrast to studies demonstrating increased cardiovascular morbidity, our pilot study did not detect significant occult atherosclerosis using MDCT of the coronaries and aorta of patients assessed five or more years following radiation for treatment of breast cancer.

Study of Estrogen Receptor and Progesterone Receptor Expression in Breast Ductal Carcinoma In Situ by Immunohistochemical Staining in ER/PgR-Negative Invasive Breast Cancer.

Background. To our knowledge, the hormone receptor status of noncontiguous ductal carcinoma in situ (DCIS) occurring concurrently in ER/PgR-negative invasive cancer has not been studied. The current study was undertaken to investigate the ER/PgR receptor status of DCIS of the breast in patients with ER/PgR-negative invasive breast cancer. Methods. We reviewed the immunohistochemical (IHC) staining for ER and PgR of 187 consecutive cases of ER/PgR-negative invasive breast cancers, collected from 1995 to 2002. To meet the criteria for the study, we evaluated ER/PgR expression of DCIS cancer outside of the invasive breast cancer. Results. A total of 37 cases of DCIS meeting the above criteria were identified. Of these, 16 cases (43.2%) showed positive staining for ER, PgR, or both. Conclusions. In our study of ER/PgR-negative invasive breast cancer we found that in 8% of cases noncontiguous ER/PR-positive DCIS was present. In light of this finding, it may be important for pathologists to evaluate the ER/PgR status of DCIS occurring in the presence of ER/PgR-negative invasive cancer, as this subgroup could be considered for chemoprevention.

Use of admission troponin in critically ill medical patients.

Serum troponin I (TnI) is a sensitive marker of cardiac injury. A relation between elevated TnI and mortality has been suggested. In this retrospective chart review of 221 patients admitted to the medical intensive care unit (MICU) during a 6-month period, the authors studied the use of admission TnI levels in predicting mortality in MICU-admitted patients. Data retrieved included demographics, admission diagnosis, troponin, electrocardiogram, Acute Physiology and Chronic Health Evaluation (APACHE) II score, echocardiogram, requirements for mechanical ventilation and vasopressor support, development of multiorgan failure, mortality, and discharge disposition. There were 132 patients for whom TnI level was sent within 24 hours of admission; these patients comprised the study group. The median age was 70 years; 59% were female. The mean APACHE II score was 22. Troponin I was positive in 31% of patients (median level, 0.4 Ug/L; range 0-358 Ug/L). The hospital mortality was 39%. Positive TnI showed a weak association with intensive care unit (ICU) mortality (P = .049) but not with overall mortality. There was no significant correlation between admission TnI concentration and APACHE II score (P = .33), administration of vasopressor medications (P = .115), or development of multiorgan failure (P = .64). The authors concluded that there is no benefit in obtaining a routine admission troponin level in MICU patients when an acute coronary event is not suspected.