-
Practical X-ray Gastric Cancer Screening Using Refined Stochastic Data Augmentation and Hard Boundary Box Training
Authors:
Hideaki Okamoto,
Takakiyo Nomura,
Kazuhito Nabeshima,
Jun Hashimoto,
Hitoshi Iyatomi
Abstract:
In gastric cancer screening, X-rays can be performed by radiographers, allowing them to see far more patients than endoscopy, which can only be performed by physicians. However, due to subsequent diagnostic difficulties, the sensitivity of gastric X-ray is only 85.5%, and little research has been done on automated diagnostic aids that directly target gastric cancer. This paper proposes a practical…
▽ More
In gastric cancer screening, X-rays can be performed by radiographers, allowing them to see far more patients than endoscopy, which can only be performed by physicians. However, due to subsequent diagnostic difficulties, the sensitivity of gastric X-ray is only 85.5%, and little research has been done on automated diagnostic aids that directly target gastric cancer. This paper proposes a practical gastric cancer screening system for X-ray images taken under realistic clinical imaging conditions. Our system not only provides a diagnostic result for each image, but also provides an explanation for the result by displaying candidate cancer areas with bounding boxes. Training object detection models to do this was very expensive in terms of assigning supervised labels, and had the disadvantage of not being able to use negative (i.e., non-cancer) data for training. Our proposal consists of two novel techniques: (1) refined stochastic gastric image augmentation (R-sGAIA) and (2) hard boundary box training (HBBT). The R-sGAIA probabilistically highlights the gastric folds in the X-ray image based on medical knowledge, thus increasing the detection efficiency of gastric cancer. The HBBT is a new, efficient, and versatile training method that can reduce the number of false positive detections by actively using negative samples. The results showed that the proposed R-sGAIA and HBBT significantly improved the F1 score by 5.9% compared to the baseline EfficientDet-D7 + RandAugment (F1: 57.8%, recall: 90.2%, precision: 42.5%). This score is higher than the physician's cancer detection rate, indicating that at least 2 out of 5 areas detected are cancerous, confirming the utility of gastric cancer screening.
△ Less
Submitted 22 March, 2023; v1 submitted 18 August, 2021;
originally announced August 2021.
-
Methods for computing $b$-functions associated with $μ$-constant deformations -- Case of inner modality 2 --
Authors:
Katsusuke Nabeshima,
Shinichi Tajima
Abstract:
New methods for computing parametric local $b$-functions are introduced for $μ$-constant deformations of semi-weighted homogeneous singularities. The keys of the methods are comprehensive Gröbner systems in Poincaré-Birkhoff-Witt algebra and holonomic ${\mathcal D}$-modules. It is shown that the use of semi-weighted homogeneity reduces the computational complexity of $b$-functions associated with…
▽ More
New methods for computing parametric local $b$-functions are introduced for $μ$-constant deformations of semi-weighted homogeneous singularities. The keys of the methods are comprehensive Gröbner systems in Poincaré-Birkhoff-Witt algebra and holonomic ${\mathcal D}$-modules. It is shown that the use of semi-weighted homogeneity reduces the computational complexity of $b$-functions associated with $μ$-constant deformations. In the case of inner modality 2, local $b$-functions associated with $μ$-constant deformations are obtained by the resulting method and given the list of parametric local $b$-functions.
△ Less
Submitted 6 January, 2021; v1 submitted 5 January, 2021;
originally announced January 2021.
-
An effective method for computing Grothendieck point residue map**s
Authors:
Shinichi Tajima,
Katsusuke Nabeshima
Abstract:
Grothendieck point residue is considered in the context of computational complex analysis. A new effective method is proposed for computing Grothendieck point residues map**s and residues. Basic ideas of our approach are the use of Grothendieck local duality and a transformation law for local cohomology classes. A new tool is devised for efficiency to solve the extended ideal membership problems…
▽ More
Grothendieck point residue is considered in the context of computational complex analysis. A new effective method is proposed for computing Grothendieck point residues map**s and residues. Basic ideas of our approach are the use of Grothendieck local duality and a transformation law for local cohomology classes. A new tool is devised for efficiency to solve the extended ideal membership problems in local rings. The resulting algorithms are described with an example to illustrate them. An extension of the proposed method to parametric cases is also discussed as an application.
△ Less
Submitted 18 November, 2020;
originally announced November 2020.
-
Computing Regular Meromorphic Differential Forms via Saito's Logarithmic Residues
Authors:
Shinichi Tajima,
Katsusuke Nabeshima
Abstract:
Logarithmic differential forms and logarithmic vector fields associated to a hypersurface with an isolated singularity are considered in the context of computational complex analysis. As applications, based on the concept of torsion differential forms due to A.G. Aleksandrov, regular meromorphic differential forms introduced by D. Barlet and M. Kersken, and Brieskorn formulae on Gauss-Manin connec…
▽ More
Logarithmic differential forms and logarithmic vector fields associated to a hypersurface with an isolated singularity are considered in the context of computational complex analysis. As applications, based on the concept of torsion differential forms due to A.G. Aleksandrov, regular meromorphic differential forms introduced by D. Barlet and M. Kersken, and Brieskorn formulae on Gauss-Manin connections are investigated. (i) A method is given to describe singular parts of regular meromorphic differential forms in terms of non-trivial logarithmic vector fields via Saito's logarithmic residues. The resulting algorithm is illustrated by using examples. (ii) A new link between Brieskorn formulae and logarithmic vector fields is discovered and an expression that rewrites Brieskorn formulae in terms of non-trivial logarithmic vector fields is presented. A new effective method is described to compute non trivial logarithmic vector fields which are suitable for the computation of Gauss-Manin connections. Some examples are given for illustration.
△ Less
Submitted 27 February, 2021; v1 submitted 20 July, 2020;
originally announced July 2020.
-
Testing zero-dimensionality of varieties at a point
Authors:
Katsusuke Nabeshima,
Shinichi Tajima
Abstract:
Effective methods are introduced for testing zero-dimensionality of varieties at a point. The motivation of this paper is to compute and analyze deformations of isolated hypersurface singularities. As an application, methods for computing local dimensions are also described. For the case where a given ideal contains parameters, the proposed algorithms can output in particular a decomposition of a…
▽ More
Effective methods are introduced for testing zero-dimensionality of varieties at a point. The motivation of this paper is to compute and analyze deformations of isolated hypersurface singularities. As an application, methods for computing local dimensions are also described. For the case where a given ideal contains parameters, the proposed algorithms can output in particular a decomposition of a parameter space into strata according to the local dimension at a point of the associated varieties. The key of the proposed algorithms is the use of the notion of comprehensive Gröbner systems.
△ Less
Submitted 29 March, 2019;
originally announced March 2019.
-
Algebraic Local Cohomology with Parameters and Parametric Standard Bases for Zero-Dimensional Ideals
Authors:
Katsusuke Nabeshima,
Shinichi Tajima
Abstract:
A computation method of algebraic local cohomology with parameters, associated with zero-dimensional ideal with parameter, is introduced. This computation method gives us in particular a decomposition of the parameter space depending on the structure of algebraic local cohomology classes. This decomposition informs us several properties of input ideals and the output of our algorithm completely de…
▽ More
A computation method of algebraic local cohomology with parameters, associated with zero-dimensional ideal with parameter, is introduced. This computation method gives us in particular a decomposition of the parameter space depending on the structure of algebraic local cohomology classes. This decomposition informs us several properties of input ideals and the output of our algorithm completely describes the multiplicity structure of input ideals. An efficient algorithm for computing a parametric standard basis of a given zero-dimensional ideal, with respect to an arbitrary local term order, is also described as an application of the computation method. The algorithm can always output "reduced" standard basis of a given zero-dimensional ideal, even if the zero-dimensional ideal has parameters.
△ Less
Submitted 27 August, 2015;
originally announced August 2015.
