TransLevelSet: Integrating Vision Transformers with Level-Sets for Medical Image Segmentation

Neurocomputing, Journal Year: 2024, Volume and Issue: 599, P. 128077 - 128077

Published: Sept. 1, 2024

Language: Английский

---

TAHIR: Transformer-Based Affine Histological Image Registration

Lecture notes in computer science, Journal Year: 2023, Volume and Issue: unknown, P. 541 - 552

Published: Jan. 1, 2023

Language: Английский

---

XDecompo: Explainable Decomposition Approach in Convolutional Neural Networks for Tumour Image Classification

Sensors, Journal Year: 2022, Volume and Issue: 22(24), P. 9875 - 9875

Published: Dec. 15, 2022

Of the various tumour types, colorectal cancer and brain tumours are still considered among most serious deadly diseases in world. Therefore, many researchers interested improving accuracy reliability of diagnostic medical machine learning models. In computer-aided diagnosis, self-supervised has been proven to be an effective solution when dealing with datasets insufficient data annotations. However, image often suffer from irregularities, making recognition task even more challenging. The class decomposition approach provided a robust such challenging problem by simplifying boundaries dataset. this paper, we propose model, called XDecompo, improve transferability features pretext downstream task. XDecompo designed based on affinity propagation-based effectively encourage explainable component highlight important pixels that contribute classification explain effect speciality extracted features. We also explore generalisability handling different datasets, as histopathology for images. quantitative results demonstrate robustness high 96.16% 94.30% CRC images, respectively. demonstrated its generalization capability achieved (both quantitatively qualitatively) compared other Moreover, post hoc method used validate feature transferability, demonstrating highly accurate representations.

Language: Английский

---

Performance Evaluation and Ranking of Deep Learning Feature Extraction Models for Thyroid Cancer Diagnosis using D-CRITIC TOPSIS

Published: Oct. 11, 2023

The nodules in the thyroid region can be cancerous or non-cancerous, present even healthy humans. Early diagnosis of cancer is helpful for prevention and treatment. Diagnosing using traditional approaches a hard-working task due to considerable burden on healthcare community. In this paper, we analyzed performance AI models (Swin Transformer, Data Efficient Image Mixer Multi-layer Perceptron) extract features from histopathological ultrasound images. Locally Linear Embedding (LLE) used reduce dimensionality feature space. These transformed are utilized training five classifiers (Random Forest classifier, Naive Bayes, Logistic Regression, Support Vector Classifier, k-nearest neighbors). There total fifteen possible combination tested 5-fold cross-validation technique, three metrics calculated. recently proposed TOPSIS technique benchmark all models, based scores, rank values assigned. distance correlation-based CRITIC (CRiteria Importance Through Intercriteria Correlation) employed weight calculation different criteria. model with Swin Transformer as extractor Random forest classifier outperformed other achieved highest score. top-ranked score 1.0000 an accuracy 0.9335 0.8518 simple deployed resource-constrained remote edge devices. With help IoT 5G/6G communication technologies, either ensemble created, federated learning techniques transfer weights order train cloud-based global model.

Language: Английский

---

Pixel-Level Clustering of Hematoxylin–Eosin-Stained Sections of Mouse and Human Biliary Tract Cancer

Biomedicines, Journal Year: 2022, Volume and Issue: 10(12), P. 3133 - 3133

Published: Dec. 5, 2022

We previously established mouse models of biliary tract cancer (BTC) based on the injection cells with epithelial stem cell properties derived from KRAS(G12V)-expressing organoids into syngeneic mice. The resulting tumors appeared to recapitulate pathological features human BTC. Here we analyzed images hematoxylin and eosin (H&E) staining for both tumor tissue cholangiocarcinoma by pixel-level clustering machine learning. A pixel-clustering model that was via training revealed homologies structure between tumors, suggesting similarities in characteristics independent animal species. Analysis samples also entropy distribution regions higher than noncancer regions, pixels thus allowing discrimination these two types regions. Histograms tended be broader late-stage cholangiocarcinoma. These analyses indicate our BTC are appropriate investigation carcinogenesis may support development new therapeutic strategies. In addition, is highly versatile contribute a diagnostic tool.

Language: Английский

---