Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

A semi-supervised Bayesian mixture modelling approach for joint batch correction and classification

bioRxiv Preprint

Systematic differences between batches of samples present significant challenges when analysing biological data. Such batch effects are well-studied and are liable to occur in any setting where multiple batches are assayed. Many existing methods for accounting for these have focused on high-dimensional data such as RNA-seq and have assumptions that reflect this. Here we focus on batch-correction in low-dimensional classification problems. We propose a semi-supervised Bayesian generative classifier based on mixture models that jointly predicts class labels and models batch effects. Our model allows observations to be probabilistically assigned to classes in a way that incorporates uncertainty arising from batch effects. By simultaneously inferring the classification and the batch-correction our method is more robust to dependence between batch and class than pre-processing steps such as ComBat. We explore two choices for the within-class densities: the multivariate normal and the multivariate t. A simulation study demonstrates that our method performs well compared to popular off-the-shelf machine learning methods and is also quick; performing 15,000 iterations on a dataset of 750 samples with 2 measurements each in 11.7 seconds for the MVN mixture model and 14.7 seconds for the MVT mixture model. We further validate our model on gene expression data where cell type (class) is known and simulate batch effects. We apply our model to two datasets generated using the enzyme-linked immunosorbent assay (ELISA), a spectrophotometric assay often used to screen for antibodies. The examples we consider were collected in 2020 and measure seropositivity for SARS-CoV-2. We use our model to estimate seroprevalence in the populations studied. We implement the models in C++ using a Metropolis-within-Gibbs algorithm, available in the R package batchmix. Scripts to recreate our analysis are at https://github.com/stcolema/BatchClassifierPaper.

Coleman, S.; Nicholls, K. C.; Castro Dopico, X.; Karlsson Hedestam, G. B.; Kirk, P. D.; Wallace, C.

2022-11-29

Radiology Radiology

Weakly Supervised Learning Significantly Reduces the Number of Labels Required for Intracranial Hemorrhage Detection on Head CT

ArXiv Preprint

Modern machine learning pipelines, in particular those based on deep learning (DL) models, require large amounts of labeled data. For classification problems, the most common learning paradigm consists of presenting labeled examples during training, thus providing strong supervision on what constitutes positive and negative samples. This constitutes a major obstacle for the development of DL models in radiology--in particular for cross-sectional imaging (e.g., computed tomography [CT] scans)--where labels must come from manual annotations by expert radiologists at the image or slice-level. These differ from examination-level annotations, which are coarser but cheaper, and could be extracted from radiology reports using natural language processing techniques. This work studies the question of what kind of labels should be collected for the problem of intracranial hemorrhage detection in brain CT. We investigate whether image-level annotations should be preferred to examination-level ones. By framing this task as a multiple instance learning problem, and employing modern attention-based DL architectures, we analyze the degree to which different levels of supervision improve detection performance. We find that strong supervision (i.e., learning with local image-level annotations) and weak supervision (i.e., learning with only global examination-level labels) achieve comparable performance in examination-level hemorrhage detection (the task of selecting the images in an examination that show signs of hemorrhage) as well as in image-level hemorrhage detection (highlighting those signs within the selected images). Furthermore, we study this behavior as a function of the number of labels available during training. Our results suggest that local labels may not be necessary at all for these tasks, drastically reducing the time and cost involved in collecting and curating datasets.

Jacopo Teneggi, Paul H. Yi, Jeremias Sulam

2022-11-29

General General

Identification of plant leaf diseases by deep learning based on channel attention and channel pruning.

In Frontiers in plant science

Plant diseases cause significant economic losses and food security in agriculture each year, with the critical path to reducing losses being accurate identification and timely diagnosis of plant diseases. Currently, deep neural networks have been extensively applied in plant disease identification, but such approaches still suffer from low identification accuracy and numerous parameters. Hence, this paper proposes a model combining channel attention and channel pruning called CACPNET, suitable for disease identification of common species. The channel attention mechanism adopts a local cross-channel strategy without dimensionality reduction, which is inserted into a ResNet-18-based model that combines global average pooling with global max pooling to effectively improve the features' extracting ability of plant leaf diseases. Based on the model's optimum feature extraction condition, unimportant channels are removed to reduce the model's parameters and complexity via the L1-norm channel weight and local compression ratio. The accuracy of CACPNET on the public dataset PlantVillage reaches 99.7% and achieves 97.7% on the local peanut leaf disease dataset. Compared with the base ResNet-18 model, the floating point operations (FLOPs) decreased by 30.35%, the parameters by 57.97%, the model size by 57.85%, and the GPU RAM requirements by 8.3%. Additionally, CACPNET outperforms current models considering inference time and throughput, reaching 22.8 ms/frame and 75.5 frames/s, respectively. The results outline that CACPNET is appealing for deployment on edge devices to improve the efficiency of precision agriculture in plant disease detection.

Chen Riyao, Qi Haixia, Liang Yu, Yang Mingchao

2022

CACPNET, channel attention, channel pruning, convolutional neural network, deep learning, plant leaf disease

General General

Rapid nondestructive detection of peanut varieties and peanut mildew based on hyperspectral imaging and stacked machine learning models.

In Frontiers in plant science

Moldy peanut seeds are damaged by mold, which seriously affects the germination rate of peanut seeds. At the same time, the quality and variety purity of peanut seeds profoundly affect the final yield of peanuts and the economic benefits of farmers. In this study, hyperspectral imaging technology was used to achieve variety classification and mold detection of peanut seeds. In addition, this paper proposed to use median filtering (MF) to preprocess hyperspectral data, use four variable selection methods to obtain characteristic wavelengths, and ensemble learning models (SEL) as a stable classification model. This paper compared the model performance of SEL and extreme gradient boosting algorithm (XGBoost), light gradient boosting algorithm (LightGBM), and type boosting algorithm (CatBoost). The results showed that the MF-LightGBM-SEL model based on hyperspectral data achieves the best performance. Its prediction accuracy on the data training and data testing reach 98.63% and 98.03%, respectively, and the modeling time was only 0.37s, which proved that the potential of the model to be used in practice. The approach of SEL combined with hyperspectral imaging techniques facilitates the development of a real-time detection system. It could perform fast and non-destructive high-precision classification of peanut seed varieties and moldy peanuts, which was of great significance for improving crop yields.

Wu Qingsong, Xu Lijia, Zou Zhiyong, Wang Jian, Zeng Qifeng, Wang Qianlong, Zhen Jiangbo, Wang Yuchao, Zhao Yongpeng, Zhou Man

2022

mildew detection, nondestructive testing, peanut seeds, stacked ensemble learning model, variety classification

General General

Stronger wind, smaller tree: Testing tree growth plasticity through a modeling approach.

In Frontiers in plant science

Plants exhibit plasticity in response to various external conditions, characterized by changes in physiological and morphological features. Although being non-negligible, compared to the other environmental factors, the effect of wind on plant growth is less extensively studied, either experimentally or computationally. This study aims to propose a modeling approach that can simulate the impact of wind on plant growth, which brings a biomechanical feedback to growth and biomass distribution into a functional-structural plant model (FSPM). Tree reaction to the wind is simulated based on the hypothesis that plants tend to fit in the environment best. This is interpreted as an optimization problem of finding the best growth-regulation sink parameter giving the maximal plant fitness (usually seed weight, but expressed as plant biomass and size). To test this hypothesis in silico, a functional-structural plant model, which simulates both the primary and secondary growth of stems, is coupled with a biomechanical model which computes forces, moments of forces, and breakage location in stems caused by both wind and self-weight increment during plant growth. The Non-dominated Sorting Genetic Algorithm II (NSGA-II) is adopted to maximize the multi-objective function (stem biomass and tree height) by determining the key parameter value controlling the biomass allocation to the secondary growth. The digital trees show considerable phenotypic plasticity under different wind speeds, whose behavior, as an emergent property, is in accordance with experimental results from works of literature: the height and leaf area of individual trees decreased with wind speed, and the diameter at the breast height (DBH) increased at low-speed wind but declined at higher-speed wind. Stronger wind results in a smaller tree. Such response of trees to the wind is realistically simulated, giving a deeper understanding of tree behavior. The result shows that the challenging task of modeling plant plasticity may be solved by optimizing the plant fitness function. Adding a biomechanical model enriches FSPMs and opens a wider application of plant models.

Wang Haoyu, Hua Jing, Kang Mengzhen, Wang Xiujuan, Fan Xing-Rong, Fourcaud Thierry, de Reffye Philippe

2022

critical wind speed, functional-structural plant model, mechanical model, optimization, thigmomorphogenesis, tree breakage

General General

EBE-YOLOv4: A lightweight detecting model for pine cones in forest.

In Frontiers in plant science

Pine cones are important forest products, and the picking process is complex. Aiming at the multi-objective and dispersed characteristics of pine cones in the forest, a machine vision detection model (EBE-YOLOV4) is designed to solve the problems of many parameters and poor computing ability of the general YOLOv4, so as to realize rapid and accurate recognition of pine cones in the forest. Taking YOLOv4 as the basic framework, this method can realize a lightweight and accurate recognition model for pine cones in forest through optimized design of the backbone and the neck networks. EfficientNet-b0 (E) is chosen as the backbone network for feature extraction to reduce parameters and improve the running speed of the model. Channel transformation BiFPN structure (B), which improves the detection rate and ensures the detection accuracy of the model, is introduced to the neck network for feature fusion. The neck network also adds a lightweight channel attention ECA-Net (E) to solve the problem of accuracy decline caused by lightweight improvement. Meanwhile, the H-Swish activation function is used to optimize the model performance to further improve the model accuracy at a small computational cost. 768 images of pine cones in forest were used as experimental data, and 1536 images were obtained after data expansion, which were divided into training set and test set at the ratio of 8:2. The CPU used in the experiment was Inter Core i9-10885@2.40Ghz, and the GPU was NVIDIA Quadro RTX 5000. The performance of YOLOv4 lightweight design was observed based on the indicators of precision (P), recall (R) and detection frames per second (FPS). The results showed that the measurement accuracy (P) of the EBE-YOLOv4 was 96.25%, the recall rate (F) was 82.72% and the detection speed (FPS) was 64.09F/S. Compared with the original YOLOv4, the precision of detection had no significant change, but the speed increased by 70%, which demonstrated the effectiveness of YOLOv4 lightweight design.

Zhang Zebing, Jiang Dapeng, Yu Huiling, Zhang Yizhuo

2022

BiFPN, ECA-Net, EfficientNet-b0, Hard-Swish, YOLOv4, pine cones detection