Yüksek dereceli gliomlarda tedaviye bağlı gelişen radyonekrozun, nüks ya da rezidü tümöral lezyondan ayrımında bilgisayar temelli yapay zekânın rolü
Yükleniyor...
Dosyalar
Tarih
2024
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Selçuk Üniversitesi, Tıp Fakültesi
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
Amaç: Gliomlar, primer malign beyin tümörlerinin büyük bir çoğunluğunu oluşturan kötü huylu tümörlerdir. Yüksek dereceli gliom vakalarında sıkça mikrovasküler proliferasyon gözlenir ve buna eşlik eden kan-beyin bariyerinin zarar görmesi post- kontrast T1 ağırlıklı manyetik rezonans görüntülemede (MRG) kontrastlanma olarak izlenir. Gelişmiş tedavi yöntemleri arasında cerrahi rezeksiyon sonrası RT ve adjuvan KT bulunmaktadır. Ancak, tedavi yanıtını değerlendirmede geleneksel MRG, özellikle tedaviye bağlı değişiklikler (TİD) olarak da adlandırılan psödoprogresyon ve radyasyon nekrozu gibi durumlarda tanıda yetersiz kalabilir. MR perfüzyon görüntülerini de içeren multiparametrik MRG kullanımıyla TİD' in radyolojik tanısında daha başarılı sonuçlar elde edilmiştir. Son yıllarda, gliom tedavi yanıtını değerlendirmek için MRG verilerini kullanan bir dizi yapay zekâ algoritması geliştirilmiştir. Bu çalışmada kemoradyoterapi (KRT) sonrası yeni gelişimli kontrastlanan lezyonu olan YDG hastalarında tedavi yanının değerlendirilmesinde denetimsiz derin öğrenme metodu ile elde edilen vasküler heterojenite ve MR perfüzyon parametrelerinin tanısal katkısını göstermeyi amaçladık. Gereç ve Yöntem: Ocak 2017- Ağustos 2023 tarihleri arasında, Selçuk Üniversitesi Tıp Fakültesi Hastanesi PACS sistemindeki kraniyal tümör protokollü MRG' ler retrospektif olarak tarandı. Çalışmaya histopatolojik tanısı Astrositom, IDH- mutant, DSÖ Grade 3- 4 olan 18 hasta ve glioblastom, IDH wild tip olan 61 hasta dahil edildi. KRT sonrası yapılan histopatolojik veya klinik-radyolojik değerlendirme sonuçlarına göre, bu hastaların 32' sine TİD tanısı, 47' sine ise rekürrens tanısı konuldu. Şüpheli lezyon tespit edilen hastaların takip MR görüntülerindeki kontrastsız ve post- kontrast T1AG, T2AG, MRP, FLAIR serileri yapay zekâ programına yüklendi. Programın hemodinamik doku işaretleme servisinde önce kontrastlanan tümör, ödem ve kontrastlanmayan nekroz alanları belirlendi. İkinci aşamada MRP verileri de kullanılarak lezyon, yüksek anjiyogenik tümör (HAT), düşük anjiyogenik tümör (DAT), infiltre periferal ödem (İPÖ) ve vazojenik periferal ödem (VPÖ) habitatlarına ayrıldı. Ardından, her bir segment ve habitat için hacimsel yüzde oranları, CBV, CBF, MTT verileri elde edildi. Verilerin rekürrens ve TİD tanıları arasındaki değişiminin istatistiksel analizi IBM SPSS Statistics 21.0 paket programında gerçekleştirildi. Sürekli sayısal değişkenlerin dağılımı Shapiro-Wilk testi ile kontrol edildi. Cinsiyet ve tedavi yanıtı sayı ve yüzde biçiminde sunuldu. Yaş ve yapay zekâ uygulamasıyla elde edilen parametrelerin değerleri ortalama ± standart sapma şeklinde gösterildi. Yapay zekâ programından elde edilen perfüzyon parametrelerinin anlamlı farklılık gösterip göstermediği Mann-Whitney U testi ile karşılaştırıldı. Receiver operating characteristic (ROC) analizi ile tanısal performans hesaplandı. ROC analizi sonuçları istatistiksel olarak anlamlı kabul edilirse Youden indeksi kullanılarak optimal kesme değerleri belirlendi. Sensitivite, spesifite, PPV, NPV ve doğruluk hesaplandı. Aksi belirtilmedikçe p<0,05 için sonuçlar istatistiksel olarak anlamlı kabul edildi. Bulgular: Çalışmanın sonuçlarına göre, HTS habitat tipleri (YAT, DAT, İPÖ, VPÖ) ve morfolojik segmentasyon alanlarında TİD ile rekürrens hastaları arasında bu alanların hacimsel yüzdelerinde ve CBV, CBF, MTT perfüzyon değerlerinde anlamlı düzeyde farklılıklar tespit edilmiştir. CBV ve CBF değerleri tüm habitatlarda, kontrastlanan tümör ve ödem alanlarında anlamlı şekilde değişiklik göstermiştir. Hacimsel oranlar, ödem alanı dışındaki diğer bölgelerde anlamlı şekilde farklılık göstermiştir. HTS habitatlarının tamamında, habitatların volumetrik (%) oranı, CBV ve CBF değerleri, TİD ve rekürrens tanısı alan hastalar arasında anlamlı olarak farklılaşmıştır. Bu değerler, kabul edilebilir, iyi ve çok iyi tanısal performans sergilemektedir. HTS habitatları arasında en yüksek AUC (0,815; %95 CI: 0,720-0,911), YAT alanının volumetrik analizinde elde edilmiştir. İntrakraniyal hacime göre YAT'ın yüzdelik oranı %0,4 ve üzerinde belirlendiğinde, %78,7 duyarlılık ve %78,1 özgüllük değerleri elde edilmiştir (p<0,001). Elde edilen tüm veriler arasında en yüksek duyarlılık, kontrastlanan tümör alanının beyin dokusuna olan oranında %83 iken, en yüksek özgüllük kontrastlanan tümör alanının CBV değerinde %96,9 olarak bulunmuştur (p<0,001). MTT parametresi İPÖ ve VPÖ habitatlarında anlamlı düzeyde farklılık göstermiştir, diğer habitatlarda anlamlı bir farklılık izlenmemiştir. Sonuç: Literatürde MRP içeren MR görüntülerle yapılan ve bazılanda yapay zekâ, derin öğrenme, CNN gibi algoritmaların kullanıldığı çok sayıda çalışma mevcuttur. Rekürrens hastalarında izlenmesi beklenen daha yüksek CBV ve CBF değerleri, elde ettiğimiz verilerle uyumluydu (p<0,05). YAT, DAT gibi anjiyogenik alanlar kullandığımız yapay zekâ yazılımı ile yapılan diğer çalışmalarda daha yüksek dereceli tümör dokusu ile ilişkili olup rekürrens hastalarında bu alanların hacimsel oranları ve CBV, CBF perfüzyon değerleri daha yüksekti (p<0,05). Farklı tümör alanlarından elde ettiğimiz sonuçlarda farklı AUC, duyarlılık ve özgüllük değerleri mevcut olup bulgular bazı çalışmalarla benzerlik göstermekte, bazı çalışmaların ulaştığı değerlerin ise gerisinde kalmaktadır. Kullandığımız yapay zekâ yazılımı, kolay kullanılabilir ve ulaşılabilir yapısı ile yapılacak güncellemeler ile birlikte gelecekte cerrahi ve KRT sonrası gliom vakalarının tedavi yanıtının değerlendirmesinde faydalı olabilecektir.
Objective: Gliomas are malignant tumors that constitute a significant majority of primary brain tumors. In cases of high-grade gliomas, microvascular proliferation is commonly observed, and the accompanying disruption of the blood-brain barrier is visualized as contrast enhancement on contrast-enhanced T1-weighted (CE-T1) magnetic resonance imaging (MRI). Advanced treatment methods include postoperative radiotherapy and adjuvant chemotherapy following surgical resection. However, traditional MRI may prove insufficient in evaluating treatment responses, particularly in cases involving treatment-related changes (TRCs) such as pseudoprogression and radiation necrosis. The use of multiparametric MRI, including MR perfusion images, has yielded more successful results. In recent years, a variety of Artificial Intelligence (AI) algorithms utilizing MRI data have been developed to assess glioma treatment responses. In this study, we aimed to demonstrate the diagnostic contribution of vascular heterogeneity and MR perfusion parameters obtained through an unsupervised deep learning method in evaluating the treatment response in patients with high grade glioma who had newly appearing contrast-enhanced lesions after chemoradiotherapy. Material and Method: Between January 1, 2017, and August 1, 2023, cranial tumor protocol MRIs within the Selçuk University Faculty of Medicine PACS system were retrospectively screened. The study included 18 patients diagnosed with grade 3-4 astrocytoma, IDH- mutant and 61 patients diagnosed with glioblastoma, IDH- wild type based on histopathological findings. Following histopathological or clinical-radiological evaluations after chemoradiotherapy, 32 patients were diagnosed with treatment-related changes (TRC), and 47 were diagnosed with progression. MR images of patients with suspected lesions, including T1-weighted post-contrast (T1AG), T2-weighted (T2AG), MR perfusion (MRP), and FLAIR sequences, were uploaded to the artificial intelligence system for further analysis. In the hemodynamic tissue signature (HTS) service of the program, contrast-enhancing tumor, edema, and non-enhancing necrosis areas were initially identified. In the second HTS process, using MRP data, the lesion was further segmented into high-angiogenic, low-angiogenic, infiltrative peripheral edema, and vasogenic peripheral edema habitats. Subsequently, volumetric percentages, as well as CBV, CBF, and MTT data, were obtained for each segment and habitat. Statistical analysis of the data's changes between progression and TRC was performed using IBM SPSS Statistics 21.0. The distribution of continuous numerical variables was assessed using the Shapiro-Wilk test. Gender and treatment response were presented in numerical and percentage format. The values of age and parameters obtained through the artificial intelligence application were presented as mean ± standard deviation. The perfusion parameters obtained from the application were compared for significant differences using the Mann-Whitney U test. Diagnostic performance was calculated using Receiver Operating Characteristic (ROC) analysis. Optimal cut-off values were determined using the Youden index if the ROC analysis results were statistically significant. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated. Unless otherwise stated, results were considered statistically significant for p < 0,05. Results: According to the results of this study, significant differences were observed in volumetric percentages and perfusion values (CBV, CBF, MTT) of HTS habitat types (HAT, LAT, IPE, VPE) and morphological segmentation areas in patients diagnosed with TRC and progression. CBV and CBF values showed significant changes in all habitats, particularly in the contrast-enhanced tumor and edema areas. Volumetric values differed significantly in areas other than the edema zone. In all HTS habitats, there were significant differences in volumetric (%) ratios, CBV, and CBF values between patients diagnosed with TRC and progression. These values demonstrate acceptable, good, and very good diagnostic performance. The highest AUC among HTS habitats was obtained in the volumetric analysis of the HAT area (AUC: 0,815; 95% CI: 0,720-0,911). When the percentage ratio of LAT to intracranial volume was determined to be 0,4% and above, sensitivity and specificity values were obtained as 78,7% and 78,1%, respectively (p<0,001). Among all obtained data, the highest sensitivity was 83% for the ratio of the contrast-enhanced tumor area to brain tissue, while the highest specificity was 96,9% for the CBV value of the contrast-enhanced tumor area (p<0,001). The MTT parameter showed significant differences in the IPE and VPE habitats, while no significant difference was observed in other habitats. Conclusion: There are numerous studies using MR images containing MRP, and in some of them, algorithms such as artificial intelligence, deep learning, CNN have been employed. The higher CBV and CBF values expected in progression patients were consistent with our findings (p<0,05). An angiogenic area like HAT and LAT, as observed in studies with Oncohabitats, was associated with higher-grade tumors, and in progression patients, the volumetric ratios and CBV, CBF perfusion values of these areas were higher (p<0,05). Our results from different tumor areas show varying AUC, sensitivity, and specificity values, with some similarities to certain studies and falling behind in values achieved by others. Oncohabitats, with its user-friendly and accessible structure, could be valuable in the future for assessing the treatment response in glioma cases post-surgery and chemoradiotherapy, particularly with upcoming updates.
Objective: Gliomas are malignant tumors that constitute a significant majority of primary brain tumors. In cases of high-grade gliomas, microvascular proliferation is commonly observed, and the accompanying disruption of the blood-brain barrier is visualized as contrast enhancement on contrast-enhanced T1-weighted (CE-T1) magnetic resonance imaging (MRI). Advanced treatment methods include postoperative radiotherapy and adjuvant chemotherapy following surgical resection. However, traditional MRI may prove insufficient in evaluating treatment responses, particularly in cases involving treatment-related changes (TRCs) such as pseudoprogression and radiation necrosis. The use of multiparametric MRI, including MR perfusion images, has yielded more successful results. In recent years, a variety of Artificial Intelligence (AI) algorithms utilizing MRI data have been developed to assess glioma treatment responses. In this study, we aimed to demonstrate the diagnostic contribution of vascular heterogeneity and MR perfusion parameters obtained through an unsupervised deep learning method in evaluating the treatment response in patients with high grade glioma who had newly appearing contrast-enhanced lesions after chemoradiotherapy. Material and Method: Between January 1, 2017, and August 1, 2023, cranial tumor protocol MRIs within the Selçuk University Faculty of Medicine PACS system were retrospectively screened. The study included 18 patients diagnosed with grade 3-4 astrocytoma, IDH- mutant and 61 patients diagnosed with glioblastoma, IDH- wild type based on histopathological findings. Following histopathological or clinical-radiological evaluations after chemoradiotherapy, 32 patients were diagnosed with treatment-related changes (TRC), and 47 were diagnosed with progression. MR images of patients with suspected lesions, including T1-weighted post-contrast (T1AG), T2-weighted (T2AG), MR perfusion (MRP), and FLAIR sequences, were uploaded to the artificial intelligence system for further analysis. In the hemodynamic tissue signature (HTS) service of the program, contrast-enhancing tumor, edema, and non-enhancing necrosis areas were initially identified. In the second HTS process, using MRP data, the lesion was further segmented into high-angiogenic, low-angiogenic, infiltrative peripheral edema, and vasogenic peripheral edema habitats. Subsequently, volumetric percentages, as well as CBV, CBF, and MTT data, were obtained for each segment and habitat. Statistical analysis of the data's changes between progression and TRC was performed using IBM SPSS Statistics 21.0. The distribution of continuous numerical variables was assessed using the Shapiro-Wilk test. Gender and treatment response were presented in numerical and percentage format. The values of age and parameters obtained through the artificial intelligence application were presented as mean ± standard deviation. The perfusion parameters obtained from the application were compared for significant differences using the Mann-Whitney U test. Diagnostic performance was calculated using Receiver Operating Characteristic (ROC) analysis. Optimal cut-off values were determined using the Youden index if the ROC analysis results were statistically significant. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated. Unless otherwise stated, results were considered statistically significant for p < 0,05. Results: According to the results of this study, significant differences were observed in volumetric percentages and perfusion values (CBV, CBF, MTT) of HTS habitat types (HAT, LAT, IPE, VPE) and morphological segmentation areas in patients diagnosed with TRC and progression. CBV and CBF values showed significant changes in all habitats, particularly in the contrast-enhanced tumor and edema areas. Volumetric values differed significantly in areas other than the edema zone. In all HTS habitats, there were significant differences in volumetric (%) ratios, CBV, and CBF values between patients diagnosed with TRC and progression. These values demonstrate acceptable, good, and very good diagnostic performance. The highest AUC among HTS habitats was obtained in the volumetric analysis of the HAT area (AUC: 0,815; 95% CI: 0,720-0,911). When the percentage ratio of LAT to intracranial volume was determined to be 0,4% and above, sensitivity and specificity values were obtained as 78,7% and 78,1%, respectively (p<0,001). Among all obtained data, the highest sensitivity was 83% for the ratio of the contrast-enhanced tumor area to brain tissue, while the highest specificity was 96,9% for the CBV value of the contrast-enhanced tumor area (p<0,001). The MTT parameter showed significant differences in the IPE and VPE habitats, while no significant difference was observed in other habitats. Conclusion: There are numerous studies using MR images containing MRP, and in some of them, algorithms such as artificial intelligence, deep learning, CNN have been employed. The higher CBV and CBF values expected in progression patients were consistent with our findings (p<0,05). An angiogenic area like HAT and LAT, as observed in studies with Oncohabitats, was associated with higher-grade tumors, and in progression patients, the volumetric ratios and CBV, CBF perfusion values of these areas were higher (p<0,05). Our results from different tumor areas show varying AUC, sensitivity, and specificity values, with some similarities to certain studies and falling behind in values achieved by others. Oncohabitats, with its user-friendly and accessible structure, could be valuable in the future for assessing the treatment response in glioma cases post-surgery and chemoradiotherapy, particularly with upcoming updates.
Açıklama
Anahtar Kelimeler
Glial Tümörler, Tedavi İlişkili Değişiklik, Tümör Rekürrensi, Yapay Zekâ, Derin Öğrenme, Perfüzyon, Glial Tumors, Treatment-Related Changes, Tumor Recurrence, Artificial Intelligence, Deep Learning, Perfusion
Kaynak
WoS Q Değeri
Scopus Q Değeri
Cilt
Sayı
Künye
Altındaş, İ. (2024). Yüksek dereceli gliomlarda tedaviye bağlı gelişen radyonekrozun, nüks ya da rezidü tümöral lezyondan ayrımında bilgisayar temelli yapay zekânın rolü. (Uzmanlık Tezi). Selçuk Üniversitesi, Tıp Fakültesi, Konya.
