MeasureNet: Measurement Based Celiac Disease Identification

Aayush Kumar Tyagi, Vaibhav Mishra, Ashok Tiwari, Lalita Mehra, Prasenjit Das, Govind Makharia, Prathosh AP, Mausam · 2024

Read MeasureNet: Measurement Based Celiac Disease Identification by Aayush Kumar Tyagi, Vaibhav Mishra, Ashok Tiwari, Lalita Mehra, Prasenjit Das, Govind Makharia, Prathosh AP, Mausam (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-01182.txt.

Introduction

Aayush Kumar Tyagi and co-authors present MeasureNet, a tool that grades celiac disease from biopsy images of the small intestine. The disease is triggered by gluten. It damages the villi, the small finger-like parts that absorb nutrients. Doctors judge severity from the ratio of villus length to crypt depth. The crypts sit at the base of the villi. Measuring that ratio by hand is slow and varies between readers.

MeasureNet finds the lines to be measured and learns from the shapes of the damage itself. A second model gives guidance on crypt location when crypts are partly hidden. A mask feature mixup technique keeps the tool from relying too heavily on that guidance. The team also built a new dataset of 750 annotated biopsy images. The tool reached about 83 percent accuracy on the two-class grade. It reached 81 percent on the full grading scale.

Catalog record

CREATORAayush Kumar Tyagi, Vaibhav Mishra, Ashok Tiwari, Lalita Mehra, Prasenjit Das, Govind Makharia, Prathosh AP, Mausam
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-01182

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

MeasureNet: Measurement Based Celiac Disease Identification Aayush Kumar Tyagi, Vaibhav Mishra, Ashok Tiwari, Lalita Mehra, Prasenjit Das, Govind Makharia, Prathosh AP, Mausam Abstract Celiac disease is an autoimmune disorder triggered by the consumption of gluten. It causes damage to the villi, the finger-like projections in the small intestine that are responsible for nutrient absorption. Additionally, the crypts, which form the base of the villi, are also affected, impairing the regenerative process. The deterioration in villi length, computed as the villi-to-crypt length ratio, indicates the severity of celiac disease. However, manual measurement of villi-crypt length can be both time-consuming and susceptible to inter-observer variability, leading to inconsistencies in diagnosis. While some methods can perform measurement as a post-hoc process, they are prone to errors in the initial stages. This gap underscores the need for pathologically driven solutions that enhance measurement accuracy and reduce human error in celiac disease assessments. Our proposed method, MeasureNet, is a pathologically driven polyline detection framework incorporating polyline localization and object-driven losses specifically designed for measurement tasks. Furthermore, we leverage segmentation model to provide auxiliary guidance about crypt location when crypt are partially visible. To ensure that model is not overdependent on segmentation mask we enhance model robustness through a mask feature mixup technique. Additionally, we introduce a novel dataset for grading celiac disease, consisting of 750 annotated duodenum biopsy images. MeasureNet achieves an 82.66% classification accuracy for binary classification and 81% accuracy for multi-class grading of celiac disease. Code: https://github.com/dair-iitd/MeasureNet 1 Introduction Bounding Box Segmentation Polyline Figure 1 : Challenges in length measurement: Measuring length along the diagonal or major axis of bounding box can overlook the curvature of the villi, leading to inaccurate assessments. Segmentation masks can produce discontinuous predictions, resulting in errors in length computation. In contrast, polyline-based measurement provides a more accurate length estimate by adapting to the curvature of the villi, effectively capturing true structure. Celiac Disease (CeD) is an autoimmune disorder triggered by consuming gluten, a protein found in wheat, barley, and rye, affecting approximately 1 in 100 people worldwide [ 32 ] . Duodenal biopsy is a primary diagnostic tool for identifying histological changes in CeD patients. Villous atrophy, characterized by the deterioration of nutrient-absorbing villi, can lead to serious complications such as malnutrition, which can be fatal in severe cases. Pathologists assess CeD using the Q-histology score as the primary indicator [ 31 ] and further evaluate disease severity by calculating the villi-to-crypt length ratio [ 7 ] . However, this measurement process is labor-intensive and prone to significant inter- and intra-observer variability [ 6 , 8 ] . Additionally, length measurement has applications beyond CeD, including assessing tumor size via the length-to-width ratio [ 28 ] and evaluating lower limb alignment [ 17 ] . These needs highlights the need of a model that ensures accurate and reliable measurement. To compute the villi-to-crypt length ratio, precise length measurement is essential. The measurement problem can be framed as a segmentation task [ 17 ] , where images are processed to generate segmentation maps of object (villi-crypt). The measurement process then becomes a post-hoc task, involving contour identification followed by quantification. However, segmentation models [ 34 , 2 ] struggle with generating consistent segmentation masks, leading to discontinuous predictions resulting in post-processing errors. Detection-based approaches, such as lane or line segment detection models designed for natural images, also perform poorly on duodenal biopsy images, where villi-crypt vary in orientation and boundaries are often indistinct (see Fig 2 ). Due to the non-trivial nature of the problem, accuracy in measurement can be improved by integrating pathological insights—such as recognizing that crypts lie between the villi shoulder and the crypt border, as shown in Fig 2 . However, a key challenge remains in embedding these expert insights within detection model. Figure 2 : Duodenal biopsy of celiac patients, indicating Villi (blue) and Crypt (yellow). Since crypts are not always fully visible, an approximate crypt depth can be estimated as distance between villi shoulder (green) and crypt border (red). We represent each villi or crypt as a polyline — a line segment composed of multiple points—and frame the task as a polyline detection problem. We propose MeasureNet , a polyline detection framework which is build on DINO-DEtection TRansformer (DINO-DETR) [ 38 ] . MeasureNet leverage general purpose transformer architecture and enhance the performance for detecting fine-grained geometric structures like polylines. Specifically, to CeD, where crypts are not clearly visible, we make approximate inference from the global attributes like villi shoulder and crypt border (see Fig 2 ). For doing so, we use segmentation model like Segformer [ 34 ] . We provide the predicted segmentation mask from segmentation model as auxiliary information for detection model to learn from it. Further, to ensure that detection model is do not suffer from exposure bias [ 1 ] , i.e. not being exposed to mistakes during training time, we augment the segmentation mask features using mixup between strong and weak segmentation mask. In addition to our method, we contribute a novel dataset, Celiac disease Detection and grading through measurement (CeDeM). CeDeM consist of 750 biopsy images of the human duodenum which have a total of 6800 polyline annotations of Villi and Crypts. Additionally, each image is provided with annotated villi shoulder and crypt border. We evaluate the performance of MeasureNet on CeDeM for localization, measurement and CeD detection. When compared to its closest baseline, MeasureNet obtains gain of 25% in localization, 33% in measurement, 9% in binary classification and 11% in grade classification accuracy. In summary: (a) We introduce the task of CeD detection through measurement, utilizing polyline detection instead of traditional methods, and demonstrate its advantages; (b) We propose MeasureNet , a polyline detection framework that incorporates localization and object-driven losses specifically designed for accurate measurement; (c) Guided by pathological insight, we use segmentation masks as auxiliary information to improve crypt length and enhance feature fusion robustness through mixup; and (d) We present CeDeM, a novel dataset of human duodenum biopsy images, containing 6,800 annotated villi and crypts, to support research in automated celiac disease detection. 2 Related Work Figure 3 : During training, given an input image ( I ), the segmentation models S_{s} and S_{w} predict strong ( M_{s} ) and weak ( M_{w} ) segmentation masks. Features are extracted from the image ( f_{I} ) as well as from each mask ( f_{s} and f_{w} ). To mitigate over-dependence on mask features and reduce exposure bias, we apply feature mixup ( f_{m} ) before the Feature Extraction (FE) stage. The image and mixed mask features are then combined using the Feature Merger (FM) module, producing a unified image-mask feature representation. These merged features are passed through the detection model D_{\theta} , a polyline detection framework, which is trained using polyline detection and object-driven loss. During inference, MeasureNet utilizes only S_{s} , eliminating the need for feature mixup. Celiac Disease: Pathologists rely on parameters like Q-histology scoring, which reflects increased intraepithelial lymphocytes (IEL), and villi-to-crypt length ratios, which indicate villous atrophy, to support CeD diagnosis [ 7 ] . Most solutions for identifying CeD use classification approaches [ 5 , 20 , 22 ] , which are often non-interpretable and offer limited assistance

Read the complete text