“Data exchange” service offers individual users metadata transfer in several different formats. Citation formats are offered for transfers in texts as for the transfer into internet pages. Citation formats include permanent links that guarantee access to cited sources. For use are commonly structured metadata schemes : Dublin Core xml and ETUB-MS xml, local adaptation of international ETD-MS scheme intended for use in academic documents.
Export
Ivanović, Sonja, 1972-
Improving QA/QC in mammography screening and breast diagnosis in Montenegro
Unapređenje kontrole kvaliteta i osiguranja kvaliteta (QA i QC) u skriningu i dijagnostičkoj mamografiji u Crnoj Gori
PDF/A (83 pages)
Mijović, Slavoljub, 1958- (mentor)
Bosmans, Hilde (mentor)
Ciraj, Olivera (član komisije)
Šćepanović, Mara, 1961- (član komisije)
Vučeljić, Mira, 1966- (član komisije)
The leading idea for this PhD project was to contribute, in a scientifically driven way, to Quality Control (QC) of mammography systems. During the project, we obtained however both scientific and practical results. Seen the very poor level in Quality Control of diagnostic radiology in Montenegro at the start of the project, the practical contribution obtained through this PhD on National level is huge. This is of course also thanks to the Technical cooperation project with the International Atomic Energy Agency (IAEA) that provided the Clinical center of Montenegro with a complete kit for mammography QC.
A first task was to perform a QC test on every mammography system in Montenegro, to compare data with European standards and to consequently find out about the situation. Concerning patient doses we were in line with the EU protocol. Concerning image quality we should be on a higher level and aim for the European acceptable level. This situation is mostly due to inadequate image processing devices and absence of any regular maintenance, but there is a factor of lack of staff and proper training, also.
The next phase of the PhD project was performed in the University Hospital of the Catholic University of Leuven, Belgium, with my remote mentor. I learned about the details of QC tests, both for film-screen and full field digital mammography. This center has also 2 breast tomosynthesis systems for which a QC protocol was being developed. More importantly, we made plans for a first scientific paper in which we would test a new feature of the MagicMax multimeter (IBA), the dosimeter that had been provided by the IAEA. This new dosimeter provides an automated estimate of the half value layer (HVL) from just a single exposure. This characteristic was tested for 5 different anode/filter combinations and results were compared to measurements with an ionization chamber, considered gold standard.
First, our scientific contributions will be presented.
In this work a critical analysis of the methodology for evaluating the x-ray half value layer was conducted. Our analysis was performed on mammography units with different anode-filter combinations. Solid state dosimeters can be very energy sensitive and may not be reliable for an automated HVL determination, while ionization chambers are known to be largely energy independent and served as gold standard. A two-phase project was started: to investigate different methods and their associated accuracies to manually calculate the HVL and to subsequently verify the accuracy of the new solid state dosimeter. First, we showed that all manually acquired, experimental data fitted well to an exponential function (correlation coefficients were in most cases more than 99%). Next, the MATLAB curve fitting toolbox was used to assess the errors on the curve fits. The uncertainty of manual HVL determination was less than 10% (usually less than 6%). These results define the errors that can be tolerated with automated equipment. From a series of measurements for analyzing accuracies, repeatability and energy dependencies of the detectors [23], and other factors [24], we realized that they are not the critical factors in HVL determination. We decided that uncertainties of HVL should not be more than 10%.
A critical factor for the HVL is the number of experimental data. The HVL, determined from the first or last four data points, obtained with successive sheets of Aluminum, was significantly different from HVL values obtained from all eight points. In every experiment three significantly different values of the HVL were obtained: the lowest, obtained from the first four data; an intermediate value, obtained from all experimental data, and the highest value, obtained from the last four points. In some cases these differences were about 25%. We concluded that the X-ray beam is hardening by passing through aluminum sheets. The beam that is being characterized is the beam as measured with the first measurement of the series that is considered in the calculation, and that it defined as the initial measurement.
Next, because new (solid state) dosimeters allow all-in-one shot data acquisition, we used the opportunities of different mammography units in Leuven and Gent to verify whether we can trust kV measurements at all anode/filters and over the complete kV range. In other words, we studied the question: are solid state dosimeters sufficiently energy corrected for (all) beam qualities? Can we trust actual dosimeters which provide ‘readymade’ data on HVL from a single exposure? Is Robson’s approach, a method used for extrapolation of tube output and HVL from a measurement at a single tube voltage, also valid on newly introduced beam qualities?
We showed that differences between the automated, direct HVL measurements of Magic Max and manual measurements did not exceed 0.02 mm Al. Standard deviation was approximately
0.1. We concluded that the direct HVL measurements with the Magic Max are within the confidence interval of manual HVL determinations.
Likewise, we used a linear model and more MATLAB software to show that Robson’s model shows a satisfactory agreement between measured and predicted values of air kerma and HVL for the conditions that we have tested.
To derive useful information from a mammogram, image processing plays important role. The x-ray beam that exits the patient contains the ‘available information’. However, the imaging system, with its specific image formation part, deteriorates this information to some extent. Part of the information content could be restored or enhanced. The presence of non-avoidable (quantum) noise makes this procedure more complex from a mathematical point of view. We have explored the use of the Wiener filter to restore the information content via ‘restoration by deconvolution’ [50]. MATLAB software was used to restore image by deconvolution procedures, using the Wiener filter in two ways: 1. with a scalar estimate of the noise/signal power ratio (NSR), 2. with a frequency dependent estimate of the noise/power ratio. In order to proof the concept, the bar-pattern was used, instead of a real breast. It is found that with both approaches, the spatial resolution is improved, although the best result is depicted in methodology 2. where a frequency dependent estimate of the noise/signal power ratio is done via respective autocorrelation functions. The conclusion was that a careful estimation of the quality of the imaging system through PSF (Point Spread Function) or MTF (Modulation Transfer Function), together with an estimation of the sort and magnitude of noise in the image, could be used to improve the spatial resolution.
A first practical contribution in the frame of radiation safety is that for the first time, Average Glandular Doses (AGD) have been measured in Montenegro. Dose reference levels were established and it is in accordance with EU protocols. We performed sensitometry and densitometry measurements at all screen-film mammography systems and they were shown to have low image quality in general. The optimization process is demanding. Optimization of image quality is already done at Podgorica’s Health care center. During the QC procedures on screen-film mammographs, we found discrepancy between the indicated and measured tube voltages, with deviations that are larger than what is allowed following the European protocol (namely ±2 kV) on the Planmed Sophie Classic.
Paper 4., 5. and 6. from the list of published papers reflect that work performed in the frame of present PhD project.
List of published papers:
1. "Dose estimations for persons occupationally exposed to ionizing radiation in Montenegro", published in Arch. of oncology 2008, 16(1-2) 5-6, A.Milatovic, S.Ivanovic, S.Jovanovic
2. Radiation protection of patients in diagnostic radiology: Status of practice in five Eastern-European countries, based on IAEA project; O. Ciraj-Bjelac, A. Beganovic, D. Faj, V. Gershand, S. Ivanovice, 1. R. Videnovicf, M. M. Rehani; European Journal of Radiology 79 (20 II) e70-e73,
3. Status of radiation protection in interventional cardiology in five Eastern -Europian Countries, O. Ciraj-Bjelac, A. Beganovicb, D. Faj , S. Ivanovice,I. R. Videnovicf, M. M. Rehani; Book of abstracts of International Conference on Radiation Protection in Medicine, Varna, Bulgaria, 1-3 September, 2010, Radiologija, Suppl 10, P 22.
4. "Getting started with protocol for Quality Assurance of Digital Mammography in the Clinical centre of Montenegro", published in Radiation Protection Dosimetry 2015, S.Ivanovic, H.Bosmans, S.Mijovic
5. Uncertainties in half value layer determination in mammography, S. Mijovic, S.Ivanovic, F. Bemelmans, H.Bosmans, AIP Conference Proceedings 1722, 300004 (2016); doi: 10.1063/1.4944308 View online: http://dx.doi.org/10.1063/1.4944308
6. Mammograms restoration by using Wiener filter, M. Dakovic, S. Ivanović, and S. Mijovic, AIP Conference Proceedings 1722, 300005 (2016); doi: 10.1063/1.4944309 View online: http://dx.doi.org/10.1063/1.4944309
Osnovna ideja na početku izrade ove doktorske teze bilo je da se kontrola kvaliteta mamografske opreme unaprijedi u naučnom smislu. Međutim, kako su odmicala mjerenja, analize i zaključci, postalo je jasno da će ova teza imati i tehnički, odnosno praktični doprinos. S obzirom da je kontrola kvaliteta u dijagnostičkoj radiologiji u Crnoj Gori na početku ovog projekta bila na nezadovoljavajućem nivou, praktični doprinos ove teze je ogroman i na nacionalnom nivou. Naravno, ovo je omogućeno zahvaljujući pomoći IAEA (Međunarodna Agencija za atomsku energiju) koja je donirala opremu za kontrolu kvaliteta.
Prvi zadatak ove teze bio je odraditi kompletnu kontrolu kvaliteta svih mamografa u zemlji, uporediti rezultate sa evropskim standardima i utvrditi naš kvalitet u odnosu na njih. Pacijentne doze naših mamografa su u okviru dozvoljenih i preporučenih doza po EU standardima. Kada govorimo o kvalitetu dijagnostičke slike, on bi trebao da bude na mnogo većem nivou od trenutnog. Ovakva situacija je najviše zbog neadekvatne opreme za razvijanje filmova, neredovnog održavanja iste, ali i zbog ljudskog faktora.
Sledeća faza je bio boravak u Univerzitetskoj katoličkoj bolnici u Luvenu, Belgija, kod mog vanjskog mentora, gdje sam u detalje naučila izvođenje QC testova za analogne i digitalne mamografe. Ovaj centar ima i 2 mamografska sistema sa tomosintezom za koje je upravo tu razvijen QC protokol. Najvažnije je bilo što je tu razvijen plan za prvi naučni rad u kome je testirana validnost MagicMax multimetra, mjerne opreme koju je Klinički centar dobio od IAEA, a koja je u tom momentu bila nova na tržištu. Ovaj novi dozimetar omogućuje direktnu procjenu podatka o debljini polusloja slabljenja snopa (HVL) iz samo jedne ekspozicije. Ova karakteristika je testirana za pet različitih anoda/filter kombinacija i rezultati su upoređeni sa jonizacionom komorom koja se smatra zlatnim standardom za ova mjerenja.
Prvo će biti objašnjen naučni doprinos ovog rada.
U ovom radu sprovedena je kritička analiza metodologije debljine sloja poluslabljenja snopa X-zraka. Različiti mamografski uređaji sa različitim anoda-filter kombinacijama uz tri različita mjerna uređaja. Čvrsti detektori su energetski veoma zavisni i mogu biti nepouzdani za automasko mjerenje HVL dok je jonizaciona komora veoma energetski nezavisna i zato služi kao zlatni stndard. Prvo smo pokazali da svi eksperimentalni podaci odgovaraju eksponencijalnoj funkciji (koeficijent korelacije je najčešće bio iznad 99%). Kao drugo smo odredili nesigurnost određivanja HVL. Alatom u MATLAB softveru procijenili smo grešku u slaganju krivih. Nesigurnost određivanja HVL je manja od 10% (najčešće manja od 6%) (Slika 3.2). Ovi rezultati su dali grešku koja se može tolerisati u slučaju direktnog mjerenja. Kao sledeće, posle velikog broja mjerenja, analiziranja tačnosti, ponovljivosti i energetske zavisnosti detektora [23], i ostalih faktora [24], zaključeno je da oni nisu kritični faktori u određivanju HVL. S obzirom da se HVL dobija relativnim mjerenjima, on nije funkcija parametra a iz jednačine 3.1. iz čega slijedi da nesigurnost određivanj a HVL ne može biti veća od 10%. Kao treće, dokazano je da je broj eksperimentalnih podataka factor koji najviše utiče na HVL. HVL koji se odredi od prvih četiri mjerenja ili zadnjih četiri, bitno se razlikuje od HVL koji se odredi iz svih osam mjerenja. U svakom obavljenom eksperimentu tri bitno različite vrijednosti HVL se dobiju: najniža vrijednost za prva četiri mjerenja, srednja vrijednost, dobijena od ukupnog broja mjerenja i najveća vrijednost dobijena od zadnja četiri mjerenja. HVL je visoko zavistan od broja izvršenih mjerenja i može se bitno razlikovati u zavisnosti od kojih se podataka računa, prva četiri mjerenja ili zadnja (npr. 0.35mm and 0.42mm). U nekim slučajevima ove razlike su bile oko 25%. Zaključak je da se ove razlike javljaju kao efekat „otvrđavanja” snopa usled prolaska kroz listove aluminijuma.
Sledeće, zbog novih ,,solid state’’ dozimetara koji omogućavaju dobijanje svih relevantnih podataka sa jednom ekspozicijom, iskorišćena je mogućnost provjere mjerenja kV za sve anoda-filter kombinacije u čitavom opsegu kV, zatim da li se može vjerovati novim dozimetrima koji određuju vrijednost HVL iz samo jedne ekspozicije, kao i validnost Robsonovog modela koji koristi metodu ektrapolacije mjerenih vrijednosti izlaza cijevi i HVL na samo jednoj vrijednosti napona. Pokazano je da odstupanje direktnog mjerenja HVL sa MagicMax multmetrom od određivanja MATLAB krivom u svakom našem slučaju ne prelazi razliku od 0.02 mm Al. Procjena standardne devijacije daje vrijednost 0.1 tako da se zaključuje da direkno očitavanje HVL sa MagicMax multimetrom unutar opsega greške određivanja HVL.
Takođe, koristeći linearni model i MATLAB softver pokazano je da eksperimentalna verifikacija Robsonovog modela (za ovu studiju odabranih kombinacija parametara) daje zadovoljavajuće slaganje između mjerenih i predviđenih vrijednosti kerme i HVL.
Da bi dobili korisnu informaciju od mamografa , obrada slike ima važnu ulogu. Uticaj sistema za dobijanje slike u toku njenog formiranja obično kvari „uhvaćenu” sliku i potrebna joj je restauracija. Prisustvo neizbežnog šuma čini ovu procedure komplikovanom sa matematičkog stanovišta. Za restauraciju mamografskih slika dekonvolucijom [50] korišćen je Viner filter kao najobjektivniji. U ovu svrhu je korišćen MATLAB softver I to na dva načina: 1. Skalarnom procjenom jačine šum/signal odnosa (NSR), 2. Procjenom zavisnosti frekvencije ovog odnosa. Umjesto pacijentkinje, korišćen je bar-patern. Nađeno je da je spacijalna rezolucija obje slike poboljšana, iako je bolji rezultat dobijen metodom 2. gdje je procjena zavisnosti frekvencije jačine odnosa šum/signal urađena respektivnom autokorelacionom funkcijom. Zaključeno je da se pažljivom procjenom imidžing sistema pomoću PSF (Point Spread Function) ili MTF (Modulation Transfer Function), zajedno sa procjenom vrste i magnitude šuma u slici spacij alna rezolucija značajno poboljšati.
U smislu tehničkog (praktičnog) doprinosa, u Crnoj Gori je u toku ove teze prvi put urađena procjena pacijentne doze (AGD - prosječna glandularna doza). Uspostavljeni su nacionalni nivoi doza koji su u skladu sa preporukama EU. Urađeni su senzitometrijski testovi i denzitometrija za sve mamografe u zemlji i ustanovljen loš kvalitet slike, generalno. Optimizacija je neophodna.
Takođe, tokom sprovođenja QC procedura na analognim mamografima pronađeno je neslaganje oko tačnosti napona cijevi između EU protokola (± 1 kV) i tehničkih uputstava proizvođača najzastupljenijeg mamografa u zemlji - Planmed Sophie Classic (±2 kV).
Radovi sa liste publikovanih radova pod 4, 5 i 6 su publikovani u toku doktorskih studija.
Lista publikovanih radova:
1. "A dose estimation for person occupationally exposed to ionizing radiation in Montenegro", published in Arch. of oncology 2008, 16( 1-2)5-6,A.Milatovic, S.Ivanovic, S.Jovanovic
2. Radiation protection of patients in diagnostic radiology: Status of practice in five Eastern-European countries, based on IAEA project; O. Ciraj-Bjelac, A. Beganovic, D. Faj, V. Gershand, S. Ivanovice, 1. R. Videnovicf, M. M. Rehanig; European Journal of Radiology 79 (20 II) e70-e73,
3. Status of radiation protection in interventional cardiology in five Eastern -Europian Countries, O. Ciraj-Bjelac, A. Beganovicb, D. Faj , S. Ivanovice,I. R. Videnovicf, M. M. Rehani; Book of abstracts of International Conference on Radiation Protection in Medicine, Varna, Bulgaria, 1-3 September, 2010, Radiologija, Suppl 10, P 22.
4. "Getting started with protocol for Quality Assurance of Digital Mammography in the Clinical centre of Montenegro", published in Radiation Protection Dosimetry 2015, S.Ivanovic, H.Bosmans, S. Mijovic
5. Uncertainties in half value layer determination in mammography, S. Mijovic, S.Ivanovic, F. Bemelmans, H.Bosmans, AIP Conference Proceedings 1722, 300004 (2016); doi: 10.1063/1.4944308 View online: http://dx.doi.org/10.1063/1.4944308
6. Mammograms restoration by using Wiener filter, M. Dakovic, S. Ivanović, and S. Mijovic, AIP Conference Proceedings 1722, 300005 (2016); doi: 10.1063/1.4944309 View online: http://dx.doi.org/10.1063/1.4944309
mammography, breast diagnosis, screening, quality control and quality assurance, X-ray diagnostic, beam quality, screen-film mammography, digital breast tomosynthesis, HVL
mamografija, dijagnostika dojke, skrining, kontrola kvaliteta i osiguranje kvaliteta, dijagnostička radiologija, kvalitet snopa, analogna mamografija, tomosinteza i digitalna mamografija, poludebljina sloja
539:615.849(043.3)
Serbian
20589060
Tekst.
The leading idea for this PhD project was to contribute, in a scientifically driven way, to Quality Control (QC) of mammography systems. During the project, we obtained however both scientific and practical results. Seen the very poor level in Quality Control of diagnostic radiology in Montenegro at the start of the project, the practical contribution obtained through this PhD on National level is huge. This is of course also thanks to the Technical cooperation project with the International Atomic Energy Agency (IAEA) that provided the Clinical center of Montenegro with a complete kit for mammography QC.
A first task was to perform a QC test on every mammography system in Montenegro, to compare data with European standards and to consequently find out about the situation. Concerning patient doses we were in line with the EU protocol. Concerning image quality we should be on a higher level and aim for the European acceptable level. This situation is mostly due to inadequate image processing devices and absence of any regular maintenance, but there is a factor of lack of staff and proper training, also.
The next phase of the PhD project was performed in the University Hospital of the Catholic University of Leuven, Belgium, with my remote mentor. I learned about the details of QC tests, both for film-screen and full field digital mammography. This center has also 2 breast tomosynthesis systems for which a QC protocol was being developed. More importantly, we made plans for a first scientific paper in which we would test a new feature of the MagicMax multimeter (IBA), the dosimeter that had been provided by the IAEA. This new dosimeter provides an automated estimate of the half value layer (HVL) from just a single exposure. This characteristic was tested for 5 different anode/filter combinations and results were compared to measurements with an ionization chamber, considered gold standard.
First, our scientific contributions will be presented.
In this work a critical analysis of the methodology for evaluating the x-ray half value layer was conducted. Our analysis was performed on mammography units with different anode-filter combinations. Solid state dosimeters can be very energy sensitive and may not be reliable for an automated HVL determination, while ionization chambers are known to be largely energy independent and served as gold standard. A two-phase project was started: to investigate different methods and their associated accuracies to manually calculate the HVL and to subsequently verify the accuracy of the new solid state dosimeter. First, we showed that all manually acquired, experimental data fitted well to an exponential function (correlation coefficients were in most cases more than 99%). Next, the MATLAB curve fitting toolbox was used to assess the errors on the curve fits. The uncertainty of manual HVL determination was less than 10% (usually less than 6%). These results define the errors that can be tolerated with automated equipment. From a series of measurements for analyzing accuracies, repeatability and energy dependencies of the detectors [23], and other factors [24], we realized that they are not the critical factors in HVL determination. We decided that uncertainties of HVL should not be more than 10%.
A critical factor for the HVL is the number of experimental data. The HVL, determined from the first or last four data points, obtained with successive sheets of Aluminum, was significantly different from HVL values obtained from all eight points. In every experiment three significantly different values of the HVL were obtained: the lowest, obtained from the first four data; an intermediate value, obtained from all experimental data, and the highest value, obtained from the last four points. In some cases these differences were about 25%. We concluded that the X-ray beam is hardening by passing through aluminum sheets. The beam that is being characterized is the beam as measured with the first measurement of the series that is considered in the calculation, and that it defined as the initial measurement.
Next, because new (solid state) dosimeters allow all-in-one shot data acquisition, we used the opportunities of different mammography units in Leuven and Gent to verify whether we can trust kV measurements at all anode/filters and over the complete kV range. In other words, we studied the question: are solid state dosimeters sufficiently energy corrected for (all) beam qualities? Can we trust actual dosimeters which provide ‘readymade’ data on HVL from a single exposure? Is Robson’s approach, a method used for extrapolation of tube output and HVL from a measurement at a single tube voltage, also valid on newly introduced beam qualities?
We showed that differences between the automated, direct HVL measurements of Magic Max and manual measurements did not exceed 0.02 mm Al. Standard deviation was approximately
0.1. We concluded that the direct HVL measurements with the Magic Max are within the confidence interval of manual HVL determinations.
Likewise, we used a linear model and more MATLAB software to show that Robson’s model shows a satisfactory agreement between measured and predicted values of air kerma and HVL for the conditions that we have tested.
To derive useful information from a mammogram, image processing plays important role. The x-ray beam that exits the patient contains the ‘available information’. However, the imaging system, with its specific image formation part, deteriorates this information to some extent. Part of the information content could be restored or enhanced. The presence of non-avoidable (quantum) noise makes this procedure more complex from a mathematical point of view. We have explored the use of the Wiener filter to restore the information content via ‘restoration by deconvolution’ [50]. MATLAB software was used to restore image by deconvolution procedures, using the Wiener filter in two ways: 1. with a scalar estimate of the noise/signal power ratio (NSR), 2. with a frequency dependent estimate of the noise/power ratio. In order to proof the concept, the bar-pattern was used, instead of a real breast. It is found that with both approaches, the spatial resolution is improved, although the best result is depicted in methodology 2. where a frequency dependent estimate of the noise/signal power ratio is done via respective autocorrelation functions. The conclusion was that a careful estimation of the quality of the imaging system through PSF (Point Spread Function) or MTF (Modulation Transfer Function), together with an estimation of the sort and magnitude of noise in the image, could be used to improve the spatial resolution.
A first practical contribution in the frame of radiation safety is that for the first time, Average Glandular Doses (AGD) have been measured in Montenegro. Dose reference levels were established and it is in accordance with EU protocols. We performed sensitometry and densitometry measurements at all screen-film mammography systems and they were shown to have low image quality in general. The optimization process is demanding. Optimization of image quality is already done at Podgorica’s Health care center. During the QC procedures on screen-film mammographs, we found discrepancy between the indicated and measured tube voltages, with deviations that are larger than what is allowed following the European protocol (namely ±2 kV) on the Planmed Sophie Classic.
Paper 4., 5. and 6. from the list of published papers reflect that work performed in the frame of present PhD project.
List of published papers:
1. "Dose estimations for persons occupationally exposed to ionizing radiation in Montenegro", published in Arch. of oncology 2008, 16(1-2) 5-6, A.Milatovic, S.Ivanovic, S.Jovanovic
2. Radiation protection of patients in diagnostic radiology: Status of practice in five Eastern-European countries, based on IAEA project; O. Ciraj-Bjelac, A. Beganovic, D. Faj, V. Gershand, S. Ivanovice, 1. R. Videnovicf, M. M. Rehani; European Journal of Radiology 79 (20 II) e70-e73,
3. Status of radiation protection in interventional cardiology in five Eastern -Europian Countries, O. Ciraj-Bjelac, A. Beganovicb, D. Faj , S. Ivanovice,I. R. Videnovicf, M. M. Rehani; Book of abstracts of International Conference on Radiation Protection in Medicine, Varna, Bulgaria, 1-3 September, 2010, Radiologija, Suppl 10, P 22.
4. "Getting started with protocol for Quality Assurance of Digital Mammography in the Clinical centre of Montenegro", published in Radiation Protection Dosimetry 2015, S.Ivanovic, H.Bosmans, S.Mijovic
5. Uncertainties in half value layer determination in mammography, S. Mijovic, S.Ivanovic, F. Bemelmans, H.Bosmans, AIP Conference Proceedings 1722, 300004 (2016); doi: 10.1063/1.4944308 View online: http://dx.doi.org/10.1063/1.4944308
6. Mammograms restoration by using Wiener filter, M. Dakovic, S. Ivanović, and S. Mijovic, AIP Conference Proceedings 1722, 300005 (2016); doi: 10.1063/1.4944309 View online: http://dx.doi.org/10.1063/1.4944309