Automated system for classifying pancreatic ultrasound images based on the segment-by-segment spectral analysis method

Classification of ultrasound images is the prevailing tool in the diagnosis of many pancreas diseases. It takes years of experience and training for a doctor to interpret an ultrasound image. Therefore, the development of models, methods and algorithms for improving the reliability and quality of interpretation of ultrasound images through the use of specialized software tools that reduce the risk of diagnostic errors is a relevant issue. The proposed method involves the segmentation of ultrasound images into segments of prescribed size of a rectangular shape and their correlation to one of three classes: oncology, pancreatitis, indifferent class. Classification is carried out by means of "strong" and "weak" classifiers. For "weak" classifiers, the Walsh-Hadamard transform is employed in the formation of descriptors. Descriptors are calculated for three "weak" classifiers. For the first "weak" classifier, the spectral coefficients of the Walsh-Hadamard transform are used, calculated for the window of the entire segment. After that, the descriptors are calculated for other "weak" classifiers, which are windows with sizes that are two and four times smaller than the sizes of the original window. The classifier consists of three independently trained neural networks – "weak" classifiers. To combine the output data of neural networks, an averaging block over the ensemble is used. Software has been developed for classifying ultrasound images which helps to create a database for the "oncology" and "pancreatitis" class segments, determine the two-dimensional Walsh-Hadamard spectrum of ultrasound image segments, train fully connected neural networks and conduct exploratory analysis to study the relevance of two-dimensional spectral coefficients. Experimental studies on the classification of ultrasound images containing oncology and pancreatitis showed an average accuracy of oncology detection – 88.4 %, and pancreatitis – 85.7 %. Errors of the second type averaged 10.2 % when pancreatitis was detected and 5.2 % when oncology was detected. To set up and test the classifiers, real data from pancreatic ultrasound were used.

1. Săftoiu A., Vilmann P., Gorunescu F., Janssen J., Hocke M., Larsen M., Iglesias-Garcia J., Arcidiacono P., Will U., Giovannini M., Dietrich CF., Havre R., Gheorghe C., McKay C., Gheonea DI., Ciurea T. Efficacy of an artificial neural network-based approach to endoscopic ultrasound elastography in diagnosis of focal pancreatic masses. Clinical Gastroenterology and Hepatology: the Official Clinical Practice Journal of the American Gastroenterological Association. 2012;10(1):84–90. DOI: 10.1016/j.cgh.2011.09.014.

2. Ozkan M., Cakiroglu M., Kocaman O., Kurt M., Yilmaz B., Can G., Korkmaz U., Dandil E., Eksi Z. Age-based computer-aided diagnosis approach for pancreatic cancer on endoscopic ultrasound images. Endoscopic Ultrasound. 2016;5(2):101. DOI: 10.4103/2303-9027.180473.

3. Tian G., Xu D., He Y., Chai W., Deng Z., Cheng C., Jin X., Wei G., Zhao Q., Jiang T. Deep learning for real-time auxiliary diagnosis of pancreatic cancer in endoscopic ultrasonography. Frontiers in Oncology. 2022;12:973652. DOI: 10.3389/fonc.2022.973652.

4. Udriștoiu A.L., Cazacu I.M., Gruionu L.G., Gruionu G., Iacob A.V., Burtea D.E., Ungureanu B.S., Costache M.I., Constantin A., Popescu C.F., Udriștoiu Ș., Săftoiu A. Real-time computer-aided diagnosis of focal pancreatic masses from endoscopic ultrasound imaging based on a hybrid convolutional and long short-term memory neural network model. PloS one. 2021;16(6):e0251701. DOI: 10.1371/journal.pone.0251701.

5. Filist S.A., Tomakova R.A., Brezhneva A.N., Malyutina I.A., Alekseev V.A. Cellular processors in multichannel image classifiers. Radiopromyshlennost' = Radio industry. 2019;(1):45–52. (In Russ.).

6. Filist S.A., Tomakova R.A., Shatalova O.V., Kuz'min A.A., Ali Kassim K.D. Method of classification of complex structured images based on self-organizing neural network structures. Radiopromyshlennost' = Radio industry. 2016;4:57–65. (In Russ.).

7. Kudryavtsev P.S., Kuz'min A.A., Filist S.A. Development of boosting methodology for classification of chest fluorograms. Biomeditsinskaya radioelektronika = Biomedical radioelectronic. 2016;9:10–15. (In Russ.).

8. Belykh V.S., Efremov M.A., Filist S.A. Development and research of a method and algorithms for intelligent classification systems of complex structured images. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta. Seriya: Upravlenie, vychislitel'naya tekhnika, informatika. Meditsinskoe priborostroenie = Proceedings of the South-West State University. Series: Control, computer engineering, information science. medical instruments engineering. 2016;(2):12–24. (In Russ.).

9. Dyudin M.V., Kudryavtsev P.S., Podmaster'ev K.V., Filist S.A., Shatalova O.V. Mathematical models for intelligent classification systems of chest radiographs. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta. Seriya: Upravlenie, vychislitel'naya tekhnika, informatika. Meditsinskoe priborostroenie = Proceedings of the South-West State University. Series: Control, computer engineering, information science. medical instruments engineering2016;(2):94–107. (In Russ.).

10. Filist S.A., Tomakova R.A., Degtyarev S.V., Rybochkin A.F. Hybrid intelligent models for chest X-ray image segmentation. Biomedical Engineering. 2018;51(5):358–363. DOI: 10.1007/s10527-018-9748-5.

11. Dabagov A.R., Gorbunov V.A., Filist S.A., Malyutina I.A., Kondrashov D.S. An Automated System for Classification of Radiographs of the Breast. Biomedical Engineering. 2020;53(6): 425–428. DOI: 10.1007/s10527-020-09957-7.

12. Kudryavtsev P.S., Kuz'min A.A., Savinov D.Yu., Filist, S.A., Shatalova O.V. Modeling of morphological formations on chest radiographs in intelligent diagnostic systems for medical purposes. Prikaspiiskii zhurnal: upravlenie i vysokie tekhnologii = Caspian journal: management and high technologies. 2017;(3):109–120. (In Russ.).

13. Filist S.A., Ali Kassim K.D., Kuz'min A.A., Shatalova O.V., Alyab'ev E.A. Formation of a feature space for classification problems of complex structured images based on spectral windows and neural network structures. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta. Seriya: Upravlenie, vychislitel'naya tekhnika, informatika. Meditsinskoe priborostroenie = Proceedings of the South-West State University. Series: Control, computer engineering, information science. medical instruments engineering. 2016;(4):56–68. (In Russ.).

14. Filist S.A., Dabagov A.R., Tomakova R.A., Malyutina I.A., Kondrashov D.S. Method of cascade segmentation of breast radiographs. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta. Seriya: Upravlenie, vychislitel'naya tekhnika, informatika. Meditsinskoe priborostroenie = Proceedings of the South-West State University. Series: Control, computer engineering, information science. medical instruments engineering. 2019;9(1):49–61. (In Russ.).

15. Filist S.A., Dabagov A.R., Tomakova R.A., Malyutina I.A., Kondrashov D.S. Multilayer morphological operators for segmentation of complex structured raster halftone images. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta. Seriya: Upravlenie, vychislitel'naya tekhnika, informatika. Meditsinskoe priborostroenie = Proceedings of the South-West State University. Series: Control, computer engineering, information science. medical instruments engineering. 2019;9(3):44–63. (In Russ.).

16. Tomakova R.A., Filist S.A., Durakov I.V. Software for automatic classification of chest radiographs based on hybrid classifiers. Ekologiya cheloveka = Human ecology. 2018;(6):59–64. (In Russ.).