AI applications to medical images: From machine learning to deep learning

Physica Medica, Год журнала: 2021, Номер 83, С. 9 - 24

Опубликована: Март 1, 2021

Язык: Английский

---

Deep Learning Predicts Lung Cancer Treatment Response from Serial Medical Imaging

Clinical Cancer Research, Год журнала: 2019, Номер 25(11), С. 3266 - 3275

Опубликована: Апрель 22, 2019

Abstract Purpose: Tumors are continuously evolving biological systems, and medical imaging is uniquely positioned to monitor changes throughout treatment. Although qualitatively tracking lesions over space time may be trivial, the development of clinically relevant, automated radiomics methods that incorporate serial data far more challenging. In this study, we evaluated deep learning networks for predicting clinical outcomes through analyzing series CT images patients with locally advanced non–small cell lung cancer (NSCLC). Experimental Design: Dataset A consists 179 stage III NSCLC treated definitive chemoradiation, pretreatment posttreatment at 1, 3, 6 months follow-up (581 scans). Models were developed using transfer convolutional neural (CNN) recurrent (RNN), single seed-point tumor localization. Pathologic response validation was performed on dataset B, comprising 89 chemoradiation surgery (178 Results: Deep models scans significantly predictive survival cancer-specific (progression, distant metastases, local-regional recurrence). Model performance enhanced each additional scan into CNN model (e.g., 2-year overall survival: AUC = 0.74, P < 0.05). The stratified low high mortality risk groups, which associated [HR 6.16; 95% confidence interval (CI), 2.17–17.44; 0.001]. also predicted pathologic in B (P 0.016). Conclusions: We demonstrate can integrate multiple timepoints improve outcome predictions. AI-based noninvasive biomarkers have a significant impact clinic given their cost minimal requirements human input.

Язык: Английский

---

Stability of feature selection algorithm: A review

Journal of King Saud University - Computer and Information Sciences, Год журнала: 2019, Номер 34(4), С. 1060 - 1073

Опубликована: Июнь 25, 2019

Feature selection technique is a knowledge discovery tool which provides an understanding of the problem through analysis most relevant features. aims at building better classifier by listing significant features also helps in reducing computational overload. Due to existing high throughput technologies and their recent advancements are resulting dimensional data due feature being treated as handy mandatory such datasets. This actually questions interpretability stability traditional algorithms. The correlation frequently produces multiple equally optimal signatures, makes method unstable thus leading instability reduces confidence selected Stability robustness preferences it perturbation training samples. indicates reproducibility power method. High algorithm important classification accuracy when evaluating performance. In this paper, we provide overview techniques algorithm. We present some solutions can handle different source instability.

Язык: Английский

---

Characterization of PET/CT images using texture analysis: the past, the present… any future?

European Journal of Nuclear Medicine and Molecular Imaging, Год журнала: 2016, Номер 44(1), С. 151 - 165

Опубликована: Июнь 6, 2016

Язык: Английский

---

Beyond imaging: The promise of radiomics

Physica Medica, Год журнала: 2017, Номер 38, С. 122 - 139

Опубликована: Июнь 1, 2017

Язык: Английский

---

Machine and deep learning methods for radiomics

Medical Physics, Год журнала: 2020, Номер 47(5)

Опубликована: Май 1, 2020

Radiomics is an emerging area in quantitative image analysis that aims to relate large‐scale extracted imaging information clinical and biological endpoints. The development of methods along with machine learning has enabled the opportunity move data science research towards translation for more personalized cancer treatments. Accumulating evidence indeed demonstrated noninvasive advanced analytics, is, radiomics, can reveal key components tumor phenotype multiple three‐dimensional lesions at time points over beyond course treatment. These developments use CT, PET, US, MR could augment patient stratification prognostication buttressing targeted therapeutic approaches. In recent years, deep architectures have their tremendous potential segmentation, reconstruction, recognition, classification. Many powerful open‐source commercial platforms are currently available embark new areas radiomics. Quantitative research, however, complex statistical principles should be followed realize its full potential. field particular, requires a renewed focus on optimal study design/reporting practices standardization acquisition, feature calculation, rigorous forward. this article, role as major computational vehicle model building radiomics‐based signatures or classifiers, diverse applications, working principles, opportunities, radiomics will reviewed examples drawn primarily from oncology. We also address issues related common applications medical physics, such standardization, extraction, building, validation.

Язык: Английский

---

Radiomics and radiogenomics in lung cancer: A review for the clinician

Lung Cancer, Год журнала: 2017, Номер 115, С. 34 - 41

Опубликована: Ноя. 8, 2017

Язык: Английский

---

Radiomics in Brain Tumor: Image Assessment, Quantitative Feature Descriptors, and Machine-Learning Approaches

American Journal of Neuroradiology, Год журнала: 2017, Номер 39(2), С. 208 - 216

Опубликована: Окт. 5, 2017

SUMMARY:

Radiomics describes a broad set of computational methods that extract quantitative features from radiographic images. The resulting can be used to inform imaging diagnosis, prognosis, and therapy response in oncology. However, major challenges remain for methodologic developments optimize feature extraction provide rapid information flow clinical settings. Equally important, clinically useful, predictive radiomic properties must clearly linked meaningful biologic characteristics qualitative familiar radiologists. Here we use cross-disciplinary approach highlight studies radiomics. We review brain tumor radiologic (eg, interpretation) through models computer vision machine learning) novel insights. outline current image prediction strategies with different levels available classes supporting decision-making. further discuss machine-learning data opportunities advance studies.

Язык: Английский

---

Diagnostic Accuracy of a Machine-Learning Approach to Coronary Computed Tomographic Angiography–Based Fractional Flow Reserve

Circulation Cardiovascular Imaging, Год журнала: 2018, Номер 11(6)

Опубликована: Июнь 1, 2018

Background: Coronary computed tomographic angiography (CTA) is a reliable modality to detect coronary artery disease. However, CTA generally overestimates stenosis severity compared with invasive angiography, and angiographic does not necessarily imply hemodynamic relevance when fractional flow reserve (FFR) used as reference. CTA-based FFR (CT-FFR), using computational fluid dynamics (CFD), improves the correlation results but computationally demanding. More recently, new machine-learning (ML) CT-FFR algorithm has been developed based on deep learning model, which can be performed regular workstation. In this large multicenter cohort, diagnostic performance ML-based was CFD-based for detection of functionally obstructive Methods Results: At 5 centers in Europe, Asia, United States, 351 patients, including 525 vessels comparison, were included. data, evaluated Correlation between excellent ( R =0.997). (area under curve, 0.84) (0.84) outperformed visual (0.69; P <0.0001). On per-vessel basis, accuracy improved from 58% (95% confidence interval, 54%–63%) by 78% (75%–82%) CT-FFR. The per-patient 71% (66%–76%) 85% (81%–89%) adding 62 85 (73%) false-positive could correctly reclassified Conclusions: On-site ML reclassifying hemodynamically nonsignificant performs equally well

Язык: Английский

---

Somatic Mutations Drive Distinct Imaging Phenotypes in Lung Cancer

Cancer Research, Год журнала: 2017, Номер 77(14), С. 3922 - 3930

Опубликована: Июнь 1, 2017

Tumors are characterized by somatic mutations that drive biological processes ultimately reflected in tumor phenotype. With regard to radiographic phenotypes, generally unconnected through present understanding the presence of specific mutations, artificial intelligence methods can automatically quantify phenotypic characters using predefined, engineered algorithms or automatic deep-learning methods, a process also known as radiomics. Here we demonstrate how imaging phenotypes be connected an integrated analysis independent datasets 763 lung adenocarcinoma patients with mutation testing and CT image analytics. We developed radiomic signatures capable distinguishing between genotypes discovery cohort (n = 353) verified them validation 352). All significantly outperformed conventional predictors (tumor volume maximum diameter). found signature related heterogeneity successfully discriminated EGFR+ EGFR- cases (AUC 0.69). Combining this clinical model EGFR status 0.70) improved prediction accuracy 0.75). The highest performing was KRAS+ tumors 0.80) and, when combined 0.81), substantially its performance 0.86). A KRAS+/KRAS- showed significant albeit lower 0.63) did not improve predictor KRAS status. Our results argue distinct predicted This work has implications for use imaging-based biomarkers clinic, applied noninvasively, repeatedly, at low cost. Cancer Res; 77(14); 3922-30. ©2017 AACR.

Язык: Английский

---