Radiol Oncol 2004; 38(3): 165-9. Computer assisted diagnosis of benign bone tumours Milan Samardziski1, George Zafiroski1, Vesna Janevska2, Daniela Miladinova3, Žaneta Popeska4 1University Clinic for Orthopaedic Surgery, Skopje, 2Pathology Institute, Skopje, 3Institute for Pathophysiology and Nuclear Medicine Skopje, 4Faculty of Natural Sciences and Mathematics, Institute for Computer Sciences, Skopje, Macedonia Background. The aim of this study is to determine the correlation between computer-assisted diagnosis (CAD) of benign bone tumours (BBT) and their histological type. Patients and method. Altogether 120 patients were included in two groups. The retrospective group com-prised 68 patients in whom the histological type of BBT was known prior to computer analysis. The prospec-tive group comprised 52 patients in whom the histological type of BBT was unknown prior to computer analysis. Computer program was efficient and easy to use. Results. Average percent of histological type confirmed with CAD in the retrospective and prospective groups was 72.06% and 76.92%, respectively. Histological confirmation of CAD in specific BBT was 91.42% for enchondroma, 96.15% for osteoid-osteoma, and 98.08% for osteochondroma. Significantly low-er percentage of CAD confirmation of fibroma, chondromixoid fibroma, osteoclastoma, desmoplastic fibro-ma and osteobalstoma due to their adverse biological character or complex anatomic localization is under-standable. Conclusions. The results speak in favour of the assumption that computer assisted diagnosis of bone tu-mours program may improve the diagnostic accuracy of the examiner. Key words: bone neoplasms — pathology; diagnosis, computer - assisted Introduction Diagnosis and treatment of benign bone tu-mours (BBT) is a multidisciplinary task. Teams of diverse subspecialists are involved Received 8 April 2004 Accepted 6 May 2004 Correspondence to: Milan Samardziski, MSc, Clinic for Orthopaedic Surgery, Vodnjanska 17, 1000 Skopje, Macedonia; Phone: +389 02 314 7626; Fax: +389 02 3165 137; E-mail: milan_samardziski@yahoo.com in the process. Good quality plain X-rays may be most helpful in 9 of 10 cases. Bone scan, CT and MRI are additionally needed for the diagnosis, staging and decision making on the management of BBT. The diagnosis of his-tological type can be done exclusively by a pa-tohistyologist.1 In the second half of the 20th century, a digital revolution started in the USA. This led to a great advance in technology and data management. A new approach in diagnostics and decision-making process in medicine was 166 Samardziski M et al. / Computer assisted diagnosis of benign bone tumours inevitable. Warner was the pioneer in computer assisted diagnosis (CAD) of congenital heart diseases in 1961.2 Lodwick in 1963 gave his preliminary results with computer assisted diagnosis of primary bone tumors.3 Many others followed him soon after: Hall in 1971, Buzdon in 1978, Virtama in 1979, Zafiroski in 1986.4-6 Our task in this study was to determine the correlation between computer-assisted diagnosis (CAD) of benign bone tu-mours (BBT) and their histological type. Patients and methods In this study, 120 patients with BBT were in-cluded. The observation period was 7 years. The patients were treated at the Clinic for Orthopaedic Surgery in Skopje. They were di-vided in two groups. The retrospective group comprised 68 patients in whom the histologi-cal type of BBT was known prior to computer analysis. The prospective group comprised 52 patients in whom the histological type of BBT was unknown prior to computer analysis. Of the total of 120 patients, 66 were males and 54 females. The age of patients ranged from 6 to 79 years old (mean 27.4 years). Two thirds (78 patients) were in the second or third decade of their life. The follow-up was from 2 to 5 years (Table 1). Osteochondroma was diagnosed in 34.16% (41) of patients and osteoid-osteoma in 35.0% (42) of patients. Enchondroma was found in 13.33% (16) of patients and 7.5% (9) patients were diagnosed with giant cell tumours. Fibroma, desmoplastic fibroma, chondroblas-toma, chondromixoid fibroma, osteoblas-toma, lipoma and hemangioma were found in 12 patients (10.0%). Enchodromas were 3 times more frequent in female patients while osteohondromas, osteoid-osteomas and giant cell tumours were more often diagnosed in male patients (Table 1). Most of the authors are using Bayes' theorem of inverse probability as a basic tool for the mathematical model in the computer program. Thomas Bayes (1702-1761) was a minister who gave the basic mathematical values to the outcome and risk, thereby founding a scientific approach to forecasting.7 Py1 Px1 y1(1- Px5 y1) .... Pxj y1 Py1 (x1, x5...xj) = —————————————————— ? Pyk Px1 yk(1- Px5yk) .... Pxi yk all k Table 1. Patients included in the study and average follow-up Benign bone tumors Age Gender Number of % Follow-up (mean yrs) M F cases (yrs) Osteoma 30 0 2 2 1.66 4.5 Osteoid-osteoma 18.3 30 12 42 35.0 5.3 Osteoblastoma 36.5 1 1 2 1.66 5 Enchondroma 40.7 4 12 16 13.33 3.6 Osteochondroma 21.7 24 17 41 34.16 3.3 Chondrobalastoma 22 1 0 1 0.83 4 Chondromyxoid fibro. 24.5 0 1 1 0.83 3.5 Osteoclastoma (GCT) 33.8 5 4 9 7.50 4.4 Hemangioma 30 0 1 1 0.83 2 Fibroma 18.7 1 2 3 2.50 3.7 Desmoplastic fibroma 14 0 1 1 0.83 3 Lipoma 39 0 1 1 0.83 3 Mean Total Total Total Mean 27.4 66 54 120 100% 3.8 Radiol Oncol 2004; 38(3): 165-9. Samardziski M et al. / Computer assisted diagnosis of benign bone tumours 167 On y axis of probability matrix, all possible diagnoses (y1, y5,...yj) are given, on x axis, all radiological characteristics of the tumours (x1, x5,...xj) are shown. P is probability, and k is the number of possible diagnosis included in the matrix. For an absolutely correct prob-ability, indefinite number of cases are needed (i), and all variables included should be com-pletely independent. An adequate vocabulary, based on the radiographic manifestations of BBT, is required for the communication with the computer program.3 The program is capable of predict-ing 34 different histological types of primary bone tumours and tumour like lesions.6 The greatest task with CAD is to achieve a correct histological type of the BBT and to follow two basic principles: (1) the prediction of the di-agnosis must be correct in the highest possi- ble number of cases (ideally in all of them), and (2) if there is a mistake in the prediction, it must not influence further treatment of the lesion in a way that could harm the patient. In the decision-making algorithm, both princi-ples are included.8 We compare our prior experiences of radiographic manifestations of BBT with the radiographic manifestations of the new cases. The next task in the algorithm is to eliminate as many data (diagnosis) as possible during the decision-making process. In this process, the strongest criteria for eliminating or in-cluding a certain diagnosis are the radiologi-cal grade of tumour growth. Many lesions are seen only in the radiological grades of tumour growth Ia, Ib or Ic (Figure 1).4 During the analysis of the x-ray, the following data were included: age and gender, localisation of the Figure 1. (a) enchondroma in the proximal phalanx of the third finger of the hand, presented with moderate pain until the fracture occurred; (b) CT imaging of osteoid-osteoma in the proximal femur, with typical “nidus”; (c) plain radiograph of the forearm showing osteochondroma of distal radius (almost not seen in frontal plane). Radiol Oncol 2004; 38(3): 165-9. 168 Samardziski M et al. / Computer assisted diagnosis of benign bone tumours 100,00 80,00 60,00 40,00 20,00 0,00 Dirflrmei d computer assisted diagnosis of BBT In both studies 72,06 76,92 Figure 2. Percentage of confirmed computer assisted diagnosis (CAD) in the retrospective and prospective study. BBT, bone destruction, destruction of the cor-tex, periostal proliferation, tumour matrix mineralisation and size of the tumour. Results In this study CAD were compared to the final histological type of BBT. The results showed high statistical significance between the radiographic manifestations of BBT and histolog-ical type. The percentage of confirmed CAD in the retrospective study was 72.06% and in the prospective study 76.92%. There was no sta-tistically significant difference between these results (?2 = 0.36; for r =0.34) (Figure 2). The analysis of different radiographic manifestations in correlation with confirmed CAD was made on a joined number of cases from both studies (retrospective and prospec-tive); so, the results gave greater statistic sig-nificance. Highest percentage of CAD was seen in the lesions localised in the cortex of the bone (83.10%) compared to the lesions lo-calized in the bone medulla (61.36%) and oth-er localizations (60.00%). Analysed parameters showed high values of ?2 test: ?2 = 7.244455; r = 0.026723 for r < 0.05. The highest percentage of confirmed CAD in correlation with expansion of the cortex under the pressure of growing BBT showed lesions without expansion (78.89%). The highest percentage of unconfirmed CAD showed lesions with the expansion of the cor-Radiol Oncol 2004; 38(3): 165-9. Confirmation of CAD in correlation with expansion of the cortex Figure 3. Percentage of confirmed computer assisted diagnosis (CAD) in correlation with the expansion of the cortex. tex greater than 10 mm (77.78%). Analysed data revealed high statistic significance (?2 = 13.76689; r = 0.001025 for r < 0.05) (Figure 3). Size of the tumours was measured in mil-limetres of their longest diameter. Tumours were divided in the group with the confirmed CAD and the group with unconfirmed CAD. Standard error and standard deviation were higher in the group with unconfirmed CAD and average size of 41.8 mm. The values showed statistical significant difference for ?2 test -21.68123; r = 0.010638 for r < 0.005. Discussion Most of the bone tumours originate from the medullar bone, destructing it prior to the growth of the lesion in the cortex. Unfor-tunately, this is not seen until 40-50% of the medullar bone is lost. In contrast to the medullar bone, the cortex shows even slight-est destruction when appropriate x-ray pro-jection is made. Slow growing and benign bone tumours produce a sclerotic reaction of the surrounding bone.9 Analysing these ma-nifestations together with bone tumour matrix one can easily determine the radiological grade of tumour growth. Active, aggressive and malignant should be immediately treated and latent (“live me alone”) bone tumours should be regularly inspected and followed.10 Working with this program for computer-assisted diagnosis of BBT appears to be easy, understandable and can be used by relatively Samardziski M et al. / Computer assisted diagnosis of benign bone tumours 169 inexperienced examiner. The use of the program improves diagnostic accuracy signifi-cantly and results in improved patient management and cost-saving.5 CAD of BBT should be confirmed in the highest possible number of cases (ideally 100%). The average percent of confirmed CAD in retrospective study is 72.06% and in prospec-tive study is 76.92%. This is slightly lower than those in previous studies of Enneking (77.9%) and Bumbasirevic (81.2%).4,9 In our study, for some specific benign bone tumours as en-chondroma, osteochondroma and osteoid-os-teoma, the confirmation is higher than 83.33%. There was no significant influence of the examiner on the results of CAD. The analysis of the results of fibroma, chon-dromixoid fibroma, osteoclastoma, desmo-plastic fibroma and osteoblastoma and lesions localized on scapula and pelvis was inconclu-sive due to their adverse biological character, low number of cases or complexity of the analysis of the specific anatomic localization. Best results of CAD were shown when lesions were localized in the cortex, in tumours without expansion of the bone and tumours with average size of 27 mm in diameter. The results support the assumption that the computer-assisted diagnosis of bone tumours program may improve the diagnostic accuracy of the examiner. This is due to an analytic, sys-tematic and logic approach to the analysis of the radiographic manifestations of BBT. A slightly lover percentage of confirmed CAD in the retrospective versus prospective study speaks in favour of that conclusion. References 1. Sundaram M. Magnetic resonance imaging for solitary lesions of bone: when, why, how useful? J Orthop Sci 1999; 4: 384-96. 2. Warner HR, Toronto AF, Veasey LG, Stephenson RE. A mathematical approach to medical diagno-sis. Application to congenital heart disease. J Am Med Ass 1961; 177: 177-83. 3. Lodwick GS, Haun CL, Smith WE, Keller RF, Robertson ED. Computer diagnosis of primary bone tumors. A prelim.report 1963; 80: 273-5. 4. Bumbasirevic Z, Buzdon P. Jedna od mogucnosti primene kompjutera u dijagnostici kostnih tumora. Medicinska Istrazivanja 1981; 14 (Suppl 1-2): 75-7. 5. Virtama P, Katevuo K, Makela P, Makinen EO. Computer aided diagnosis of bone tumors. Acta Radiol Diagn (Stockh). 1979; 20: 70-4. 6. Zafiroski G. Maligni koskeni timori. Skopje: Studentski zbor-Skopje; 1986. p. 11-14. 7. Mercer R. The caring doctor. Chapter 1. In: Art and practice of children’s orthopedics; Raven Press, NY, 1992. 8. Ledley R S, Lusted LB. Reasoning foundations of medical diagnosis. Science 1959; 130: 9-21. 9. Enneking WF. A system of staging musculoskele-tal neoplasms. Clin Orthop 1986; 204: 9-24. 10. Simon MA, Springfield D. Surgery for bone and soft-tissue tumors. Philadelphia: Lippinkot-Raven; 1997: 119-207. Radiol Oncol 2004; 38(3): 165-9.