FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3136838961> ?p ?o ?g. }

• W3136838961 endingPage "361" @default.
• W3136838961 startingPage "359" @default.
• W3136838961 abstract "HomeRadiologyVol. 299, No. 2 PreviousNext Reviews and CommentaryFree AccessEditorialHigh-Precision Assessment of Chemoradiotherapy of Rectal Cancer with Near-Infrared Photoacoustic Microscopy and Deep LearningAlexander L. Klibanov Alexander L. Klibanov Author AffiliationsFrom the Cardiovascular Division and Robert M. Berne Cardiovascular Research Center, University of Virginia, 409 Lane Rd, UVA CVRC, PO Box 801394, Charlottesville VA 22908.Address correspondence to the author (e-mail: [email protected]).Alexander L. Klibanov Published Online:Mar 23 2021https://doi.org/10.1148/radiol.2021210261MoreSectionsPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked In See also the article by Leng and Uddin et al in this issue.Dr Klibanov is an associate professor, with a primary appointment in the Cardiovascular Division, Department of Medicine, at the University of Virginia School of Medicine. He has joint appointments in the Departments of Biomedical Engineering and Radiology at this same institution. His research interests are in the areas of molecular imaging (mostly with US) and US-triggered drug and gene delivery. He is a chemist by initial training and a graduate of the Moscow University chemistry department. He is a recipient of the Torsten Almen Award at the Contrast Media Research Symposium series.Download as PowerPointOpen in Image Viewer MRI is assumed to be the standard of reference in cancer imaging; however, most of the time alternative imaging tools are used for screening or monitoring treatment response or disease progression. There are cases where other imaging modalities, including the most innovative ones, might become the primary imaging tools applicable for a watchful waiting approach, where surgery may be avoided after chemotherapy or radiation therapy.Colorectal cancer is the third most common cancer diagnosed in the United States and worldwide. Specifically, for rectal cancer, almost 45 000 new cases are reported in the United States every year (1). Lifetime risk of rectal cancer is slightly grater than 1%. Therefore, the diagnosis of this disease and especially treatment efficacy monitoring could be important. In this issue of Radiology, Leng and Uddin et al (2) describe a successful pilot clinical trial undertaken with exactly that purpose. A cutting-edge imaging technique, photoacoustic microscopy (PAM) at the near-infrared-II wavelength, is combined with US and modern image processing by using deep learning convolutional neural networks (CNNs). The authors report an excellent ability to detect the presence or absence of cancer after chemotherapy or radiation therapy (2).The principles of photoacoustics were proposed by Alexander Graham Bell in the 19th century. When short pulses of energy are absorbed by matter, the matter will expand due to heating and will generate outgoing sound waves after each energy pulse. Biomedical application of photoacoustic imaging gradually progressed from preclinical testing to clinical research (3). There are more than 40 photoacoustic trials currently listed on https://www.clinicaltrials.gov. Hopes are expressed to achieve authorization for clinical use within a reasonable period.In modern times, nanosecond pulses of laser light serve as the energy source to achieve safe and very moderate but rapid expansion of biologic tissue. This expansion generates ultrasound waves at the megahertz frequency range. Light absorption (and resulting acoustic energy generation) is higher in the tissue with a higher concentration of dye molecules (eg, hemoglobin in red blood cells). US imaging can be adapted for detection and observation of these photoacoustic emission patterns, such as blood within the vessels, and can even be used to determine the local blood oxygenation levels by distinguishing absorption spectra of oxygenated and nonoxygenated hemoglobin (4). Spatial resolution of ultrasound is dependent on the frequency of the US probe used; in this study (2), with the 20-MHz center frequency probe, the axial resolution is listed as approximately 50 μm and lateral resolution is listed as approximately 65 μm. An added benefit is the ability to combine photoacoustics with the traditional send-and-receive US imaging. Thus, fusion images add the data on positioning within the body anatomy, which simplifies image analysis. US-only imaging data are also collected for downstream testing.A detailed description of the photoacoustic hardware and imaging setup, with two orders of magnitude slower frame rate and lower laser wavelength, has been made available earlier (5). In the current setup (2), an approved endorectal US probe is fitted with an optical fiber for kilohertz frame rate laser light illumination to assess the rectum for the presence or absence of the underlying tumor. This is achieved by investigating normal bowel layered structure and its underlying microvasculature network, as well as normalized microvasculature after tumor disappearance in case chemoradiotherapy leads to complete response. The method can also be used to distinguish normal and normalized microvasculature from the abnormal vasculature of underlying tumor (if the latter is still present as a residual disease after chemoradiotherapy). Light energy absorption by blood in the vasculature serves as an optimal natural contrast agent for photoacoustic imaging. The report (2) states that resolution of the PAM technique in this study is better than that of MRI.The design of this clinical trial takes advantage of the planned surgery after neoadjuvant chemoradiotherapy. PAM and US imaging take place immediately before surgery in the in vivo arm of the study or with the ex vivo samples of rectal tissue that are removed during surgical resection. T2- and diffusion-weighted MRI and endoscopy studies are performed. Tissue histologic results are used as the reference standard. PAM and US images for the training data set are obtained mostly ex vivo from postsurgical resected tissue material (in >20 patients). In 10 more patients, in vivo imaging is performed immediately before surgery, and data from five patients are included in the training sets. Collectively, over 2000 PAMs (over 1000 each for malignant and normal groups) and over 2500 US images are used as standard sets, with known ground truth.These PAM- and US-only images are used as training sets that are fed into the deep learning CNN. Neural network artificial intelligence is considered highly successful for analysis of medical imaging results (6), including photoacoustic findings (7). The series of PAM and US images is fed into the feature extraction and pooling layers. The resulting data matrix arrays, which have many fewer data points, become manageable and are then processed by the CNN classification layers. The final one, a fully connected output layer single-dimensional array, has only two outputs, either the normal rectal tissue or the malignant tissue, with respective probability number. Repeated CNN adjustment and training cycles on the subsets of data are then performed, with cross-validation against ground truth histologic data (clinical standard of care), where no viable tumor cells are found, or incomplete treatment, where residual tumor cells are observed.Once a CNN is established and considered trained, it is then used to analyze the separate image sets for the five test patients obtained immediately prior to surgery that have not been evaluated by the CNN earlier. The resulting prediction of the PAM CNN and US CNN is compared against ground truth of the tissue histologic investigation performed for the surgical specimen. Receiver operating characteristic curve plot for the test patient study, which provides information on true-positive PAM CNN prediction rate versus false-positive rate, is remarkably good, with an area under the receiver operating characteristic curve (AUC) of 0.98. Therefore, PAM CNN may provide a reliable recommendation whether follow-up surgery needs to be performed to remove residual tumor or prior tumor treatment has been successful with no residual tumor present. US CNN, while still better than a random AUC of 0.5, is not as successful and reliable at this prediction, with an AUC of 0.71, which is not surprising given the inability of B-mode US to image microvasculature, which PAM can do. Also, gray-scale US pattern detection that relies solely on the presence or absence of the undisturbed layered structures seems to be insufficiently distinctive to enable reliable prediction.There were several limitations to the study as described, as well as the technique itself. While the number of imaging frames observed and analyzed for PAM CNN training exceeded 2000, the number of actual patients in the test group, for whom the diagnosis was established with PAM CNN, was only five, with just over 150 region of interest images. Therefore, one may advocate expanding the clinical trial size to validate the conclusions for a larger cohort of patients. Technique limitations are also clear: PAM requires dedicated pulsed lasers, which are often large and heavy, cumbersome (eg, with water cooling), not always reliable (with the expected work time amounting to less than 300 hours of actual operation), and somewhat expensive. PAM penetration depth, although higher for the 1064-nm light, still looks relatively superficial due to tissue attenuation.Competing imaging modalities for the proposed diagnostic task are obvious and were discussed by Leng and Uddin et al (2) only briefly. One of the options to consider would be MRI with the endorectal coil; due to proximity to the tissue of interest, it will very likely improve the spatial resolution, which regular MRI with a body coil would not always provide. Another alternative to PAM, potentially easy to implement even in a smaller institution environment, is the use of contrast-enhanced US. While regular US of deeply located organs requires low frequency (single megahertz range), which leads to millimeter-level resolution, the use of blood pool US contrast microbubbles might address this deficiency (8). With the advent of super-resolution imaging, high frame rate used in the modern imaging equipment allows high-resolution imaging of centroids and trajectories of sparsely situated individual contrast particles in the bloodstream as they travel through the vasculature after intravenous administration (9). The spatial resolution of this contrast US will likely meet that for PAM. The use of an endorectal US probe (10) with higher US frequency will also improve image quality and provide information on blood flow. Combination of higher-frequency US with super-resolution contrast imaging will allow further improvement in resolution and might be able to compete with PAM CNN.Overall, the clinical study (2) reports an exciting capability for minimally invasive detection of residual or recurrent disease after chemotherapy and radiation therapy of locally advanced rectal cancer. The study demonstrates excellent positive and negative predictive values of photoacoustic microscopy when combined with cognitive neural network image analysis (area under the receiver operating curve, 0.98). Further progression of the study toward clinical adoption will provide an imaging tool that may improve the quality of life of patients after chemoradiotherapy of rectal cancer by safely avoiding unnecessary surgery.Disclosures of Conflicts of Interest: A.L.K. Activities related to the present article: institution has a subcontract from SoundPipe Therapeutics via National Institutes of Health grant R44 HL139241. Activities not related to the present article: institution holds U.S. patents 9,949722; 10,507,315; and 10,155,063 and holds U.S. patent application 20180360755 A1. Other relationships: was a cofounder and minority shareholder in Targeson, a startup in the area of preclinical targeted microbubbles, that has since dissolved.Supported in part by the National Institute of Biomedical Imaging and Bioengineering (R01EB023055), National Institute of Neurological Disorders and Stroke (R01NS076726), and National Heart, Lung, and Blood Institute (R44HL139241). The content of this publication is solely the responsibility of the author and does not necessarily represent the official views of the National Institutes of Health.References1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2019. CA Cancer J Clin 2019;69(1):7–34. Crossref, Medline, Google Scholar2. Leng X, Uddin KMS, Chapman W, Assessing rectal cancer treatment response using coregistered endorectal photoacoustic and US imaging paired with deep learning. Radiology 2021. https://doi.org/10.1148/radiol.2021202208. Published online March 23, 2021. Google Scholar3. Wang LV, Hu S. Photoacoustic tomography: in vivo imaging from organelles to organs. Science 2012;335(6075):1458–1462. Crossref, Medline, Google Scholar4. Cao R, Li J, Ning B, Functional and oxygen-metabolic photoacoustic microscopy of the awake mouse brain. Neuroimage 2017;150:77–87. Crossref, Medline, Google Scholar5. Leng X, Chapman W Jr, Rao B, Feasibility of co-registered ultrasound and acoustic-resolution photoacoustic imaging of human colorectal cancer. Biomed Opt Express 2018;9(11):5159–5172. Crossref, Medline, Google Scholar6. Bluemke DA. Radiology in 2018: Are You Working with AI or Being Replaced by AI?. Radiology 2018;287(2):365–366. Link, Google Scholar7. Yang C, Lan H, Gao F, Gao F. Review of deep learning for photoacoustic imaging. Photoacoustics 2020;21:100215. Crossref, Medline, Google Scholar8. Zhuang H, Yang ZG, Chen HJ, Peng YL, Li L. Time-intensity curve parameters in colorectal tumours measured using double contrast-enhanced ultrasound: correlations with tumour angiogenesis. Colorectal Dis 2012;14(2):181–187. Crossref, Medline, Google Scholar9. Errico C, Pierre J, Pezet S, Ultrafast ultrasound localization microscopy for deep super-resolution vascular imaging. Nature 2015;527(7579):499–502. Crossref, Medline, Google Scholar10. Bogdanova EM, Orlova LP, Trubacheva YuL, Khomyakov EA, Rybakov EG. The first experience of contrast-enhanced ultrasound in epithelial rectal tumors. Koloproctologia 2020;19(4):57–70. Crossref, Google ScholarArticle HistoryReceived: Jan 28 2021Revision requested: Feb 10 2021Revision received: Feb 12 2021Accepted: Feb 16 2021Published online: Mar 23 2021Published in print: May 2021 FiguresReferencesRelatedDetailsAccompanying This ArticleAssessing Rectal Cancer Treatment Response Using Coregistered Endorectal Photoacoustic and US Imaging Paired with Deep LearningMar 23 2021RadiologyRecommended Articles Assessing Rectal Cancer Treatment Response Using Coregistered Endorectal Photoacoustic and US Imaging Paired with Deep LearningRadiology2021Volume: 299Issue: 2pp. 349-358Using Deep Learning for MRI to Identify Responders to Chemoradiotherapy in Rectal CancerRadiology2020Volume: 296Issue: 1pp. 65-66Chimeric Antigen Receptor T-cell Immunotherapy Induces Transient Tumor Hyperoxia Instead of HypoxiaRadiology: Imaging Cancer2021Volume: 3Issue: 1Can We Use MRI and US to Predict Axillary Node Response in Breast Cancer?Radiology2019Volume: 293Issue: 1pp. 58-59Predicting Rectal Cancer Response to Neoadjuvant Chemoradiotherapy Using Deep Learning of Diffusion Kurtosis MRIRadiology2020Volume: 296Issue: 1pp. 56-64See More RSNA Education Exhibits Contrast-enhanced Ultrasound As A Problem-solving Tool: A Guide To A Successful Integration Into Clinical RoutineDigital Posters2021Posttreatment Challenges in Rectal Cancer MRI: A Systematic ApproachDigital Posters2022MRI Evaluation of Pathological Complete Response of Locally Advanced Rectal Cancer after Neoadjuvant Chemoradiation Therapy: Can We Wait and What Will We Watch?Digital Posters2019 RSNA Case Collection Secondary Angiosarcoma of the BreastRSNA Case Collection2021Arterioureteral fistulaRSNA Case Collection2021Bilateral retinoblastoma RSNA Case Collection2021 Vol. 299, No. 2 Metrics Altmetric Score PDF download" @default.
• W3136838961 created "2021-03-29" @default.
• W3136838961 creator A5077766065 @default.
• W3136838961 date "2021-05-01" @default.
• W3136838961 modified "2023-09-23" @default.
• W3136838961 title "High-Precision Assessment of Chemoradiotherapy of Rectal Cancer with Near-Infrared Photoacoustic Microscopy and Deep Learning" @default.
• W3136838961 cites W2030964111 @default.
• W3136838961 cites W2109005881 @default.
• W3136838961 cites W2174473997 @default.
• W3136838961 cites W2577529438 @default.
• W3136838961 cites W2802896687 @default.
• W3136838961 cites W2894740621 @default.
• W3136838961 cites W2911188335 @default.
• W3136838961 cites W3111805610 @default.
• W3136838961 cites W3115408395 @default.
• W3136838961 cites W3136592095 @default.
• W3136838961 doi "https://doi.org/10.1148/radiol.2021210261" @default.
• W3136838961 hasPubMedCentralId "https://www.ncbi.nlm.nih.gov/pmc/articles/8103911" @default.
• W3136838961 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/33759582" @default.
• W3136838961 hasPublicationYear "2021" @default.
• W3136838961 type Work @default.
• W3136838961 sameAs 3136838961 @default.
• W3136838961 citedByCount "0" @default.
• W3136838961 crossrefType "journal-article" @default.
• W3136838961 hasAuthorship W3136838961A5077766065 @default.
• W3136838961 hasBestOaLocation W31368389612 @default.
• W3136838961 hasConcept C120665830 @default.
• W3136838961 hasConcept C121332964 @default.
• W3136838961 hasConcept C121608353 @default.
• W3136838961 hasConcept C126322002 @default.
• W3136838961 hasConcept C126838900 @default.
• W3136838961 hasConcept C141071460 @default.
• W3136838961 hasConcept C142724271 @default.
• W3136838961 hasConcept C147080431 @default.
• W3136838961 hasConcept C2778424827 @default.
• W3136838961 hasConcept C2781074409 @default.
• W3136838961 hasConcept C509974204 @default.
• W3136838961 hasConcept C526805850 @default.
• W3136838961 hasConcept C54932901 @default.
• W3136838961 hasConcept C71924100 @default.
• W3136838961 hasConceptScore W3136838961C120665830 @default.
• W3136838961 hasConceptScore W3136838961C121332964 @default.
• W3136838961 hasConceptScore W3136838961C121608353 @default.
• W3136838961 hasConceptScore W3136838961C126322002 @default.
• W3136838961 hasConceptScore W3136838961C126838900 @default.
• W3136838961 hasConceptScore W3136838961C141071460 @default.
• W3136838961 hasConceptScore W3136838961C142724271 @default.
• W3136838961 hasConceptScore W3136838961C147080431 @default.
• W3136838961 hasConceptScore W3136838961C2778424827 @default.
• W3136838961 hasConceptScore W3136838961C2781074409 @default.
• W3136838961 hasConceptScore W3136838961C509974204 @default.
• W3136838961 hasConceptScore W3136838961C526805850 @default.
• W3136838961 hasConceptScore W3136838961C54932901 @default.
• W3136838961 hasConceptScore W3136838961C71924100 @default.
• W3136838961 hasFunder F4320337338 @default.
• W3136838961 hasFunder F4320337359 @default.
• W3136838961 hasFunder F4320337363 @default.
• W3136838961 hasIssue "2" @default.
• W3136838961 hasLocation W31368389611 @default.
• W3136838961 hasLocation W31368389612 @default.
• W3136838961 hasLocation W31368389613 @default.
• W3136838961 hasOpenAccess W3136838961 @default.
• W3136838961 hasPrimaryLocation W31368389611 @default.
• W3136838961 hasRelatedWork W1978401599 @default.
• W3136838961 hasRelatedWork W2070953787 @default.
• W3136838961 hasRelatedWork W2092105160 @default.
• W3136838961 hasRelatedWork W2092663273 @default.
• W3136838961 hasRelatedWork W2150621467 @default.
• W3136838961 hasRelatedWork W2392730113 @default.
• W3136838961 hasRelatedWork W2410827424 @default.
• W3136838961 hasRelatedWork W2413847564 @default.
• W3136838961 hasRelatedWork W3004259476 @default.
• W3136838961 hasRelatedWork W3090963952 @default.
• W3136838961 hasVolume "299" @default.
• W3136838961 isParatext "false" @default.
• W3136838961 isRetracted "false" @default.
• W3136838961 magId "3136838961" @default.
• W3136838961 workType "article" @default.