Brain Tumor Segmentation Using Convolutional Neural Networks in MRI Images

Among brain tumors, gliomas are the most common and aggressive, leading to a very short life expectancy in their highest grade. Thus, treatment planning is a key stage to improve the quality of life of oncological patients. Magnetic resonance imaging (MRI) is a widely used imaging technique to assess these tumors, but the large amount of data produced by MRI prevents manual segmentation in a reasonable time, limiting the use of precise quantitative measurements in the clinical practice. So, automatic and reliable segmentation methods are required; however, the large spatial and structural variability among brain tumors make automatic segmentation a challenging problem. In this paper, we propose an automatic segmentation method based on Convolutional Neural Networks (CNN), exploring small 3 ×3 kernels. The use of small kernels allows designing a deeper architecture, besides having a positive effect against overfitting, given the fewer number of weights in the network. We also investigated the use of intensity normalization as a pre-processing step, which though not common in CNN-based segmentation methods, proved together with data augmentation to be very effective for brain tumor segmentation in MRI images. Our proposal was validated in the Brain Tumor Segmentation Challenge 2013 database (BRATS 2013), obtaining simultaneously the first position for the complete, core, and enhancing regions in Dice Similarity Coefficient metric (0.88, 0.83, 0.77) for the Challenge data set. Also, it obtained the overall first position by the online evaluation platform. We also participated in the on-site BRATS 2015 Challenge using the same model, obtaining the second place, with Dice Similarity Coefficient metric of 0.78, 0.65, and 0.75 for the complete, core, and enhancing regions, respectively.

Assumptions in the radiotherapy of glioblastoma

In the light of advances in computerized tomography (CT), we have retrospectively evaluated the assumptions that underlie the radiation therapy of glioblastoma: (1) No neuroradiologic technique provides an accurate delineation of tumor bulk and location, (2) glioblastoma is commonly multicentric, and (3) a major source of therapeutic failure is recurrence beyond radiotherapy fields. 1. CT scans, performed on glioblastoma patients within 2 months of postmortem examination, defined both gross and microscopic tumor extent (within a 2-cm margin) in all but 6 of 35 patients evaluated. The major source of error was subependymal spread (four patients). 2. Multicentricity occurred in only 4% of untreated and 6% of treated (radiotherapy with or without chemotherapy) patients. All multicentric lesions were identified on CT scans. 3. Serial CT scans on 42 patients revealed that glioblastoma recurred within a 2-cm margin of the primary site in 90%. Occurrences outside this margin were accurately delineated by CT in all instances. Because most patients show recurrence within or in close proximity to the original site, current radiation doses would appear to be inadequate for therapy of the primary tumor. CT scan accuracy may permit smaller-field and higher-dose irradiation therapy for glioblastoma.

Radiation-induced dementia in patients cured of brain metastases

When a patient with cancer develops a brain metastasis, death is usually imminent, but aggressive treatment in some patients with limited or no systemic disease yields long-term survival. In such patients, delayed deleterious effects of therapy are particularly tragic. We report 12 patients who developed delayed complications of whole brain radiotherapy (WBRT) given as sole treatment (4 patients) or in combination with surgical resection (8 patients). Within 5 to 36 months (median, 14) all patients developed progressive dementia, ataxia, and urinary incontinence causing severe disability in all and leading to death in 7. No patient had tumor recurrence when neurologic symptoms began. Cortical atrophy and hypodense white matter were identified by CT in all. Contrast-enhancing lesions were seen in 3 patients; 2 of the lesions yielded radionecrosis on biopsy. Autopsies on 2 patients revealed diffuse chronic edema of the hemispheric white matter in the absence of tumor recurrence. Corticosteroids and ventriculoperitoneal shunt offered significant but incomplete improvement in some patients. The total dose of WBRT was only 2,500 to 3,900 cGy, but daily fractions of 300 to 600 cGy were employed. We believe that these fractionation schedules, several of which are used commonly, predispose to delayed neurologic toxicity, and that more protracted schedules should be employed for the safe and efficacious treatment of good-risk patients with brain metastases. The incidence of WBRT-induced dementia was only 1.9 to 5.1% in the 2 populations reviewed here; however, this underestimates the incidence because only severely affected patients could be identified from chart review.

Glucose utilization of cerebral gliomas measured by [18F] fluorodeoxyglucose and positron emission tomography

Positron emission tomography was used to measure local cerebral glucose utilization by the 1-[18F]fluoro-2-deoxy-D-glucose technique in 23 patients with cerebral gliomas. All 10 high-grade (III and IV) astrocytomas demonstrated a region of high activity with a glucose consumption of 7.4 +/- 3.5 (SD) mg/100 gm per minute. The 13 low-grade (I and II) gliomas had a glucose metabolic rate of 4.0 +/- 1.8 mg/100 gm per minute, with no distinctly visible hot spot. Thus, we found a correlation between rate of glycolysis and malignancy in primary cerebral tumors. Cerebral cortical glucose utilization was often depressed in areas adjacent to or neurally connected to the tumor site, and there was focal irregular delta wave EEG activity in these areas.

Patterns of failure following treatment for glioblastoma multiforme and anaplastic astrocytoma

Recurrence patterns of glioblastoma multiforme (25) and anaplastic astrocytoma (9) were studied using CT scans of 34 patients who received all or a portion of their surgical treatment at Memorial Sloan-Kettering Cancer Center from January 1983 through February 1987. Thirty-two patients presented with unifocal tumors and two with multifocal tumors. All patients received radiation therapy following initial surgery. Eighteen patients who underwent re-operation following CT evidence of recurrence had histologic verification of recurrent tumor; sixteen patients had radiographic evidence of recurrence only. Seventy-eight percent (25/32) of unifocal tumors recurred within 2.0 cm of the pre-surgical, initial tumor margin, defined as the enhancing edge of the tumor on CT scan. Fifty-six percent (18/32) of tumors recurred within 1.0 cm of the initial tumor margin. Tumors for which a gross total resection was accomplished tended to recur closer to the initial tumor margin than did subtotally resected tumors (p greater than 0.1). Extensive pre-operative edema was associated with a decreased distance between initial and recurrent tumor margins. Large tumors were generally not more likely to recur further from the initial tumor margin than were smaller tumors. No unifocal tumor recurred as a multifocal tumor. Only one tumor (initially near the midline) recurred in the contralateral hemisphere. The findings support the use of partial brain irradiation for post-operative treatment of glioblastoma multiforme and anaplastic astrocytomas, and may help to determine the most appropriate treatment volume for interstitial irradiation.

Imaging-based stereotaxic serial biopsies in untreated intracranial glial neoplasms

Forty patients with previously untreated intracranial glial neoplasms underwent stereotaxic serial biopsies assisted by computerized tomography (CT) and magnetic resonance imaging (MRI). Tumor volumes defined by computer reconstruction of contrast enhancement and low-attenuation boundaries on CT and T1 and T2 prolongation on MRI revealed that tumor volumes defined by T2-weighted MRI scans were larger than those defined by low-attenuation or contrast enhancement on CT scans. Histological analysis of 195 biopsy specimens obtained from various locations within the volumes defined by CT and MRI revealed that: contrast enhancement most often corresponded to tumor tissue without intervening parenchyma; hypodensity corresponded to parenchyma infiltrated by isolated tumor cells or in some instances to tumor tissue in low-grade gliomas or to simple edema; and isolated tumor cell infiltration extended at least as far as T2 prolongation on magnetic resonance images. This information may be useful in planning surgical procedures and radiation therapy in patients with intracranial glial neoplasms.

Dabrafenib plus trametinib in patients with BRAFV600-mutant melanoma brain metastases (COMBI-MB): a multicentre, multicohort, open-label, phase 2 trial

BACKGROUND

Dabrafenib plus trametinib improves clinical outcomes in BRAFV600-mutant metastatic melanoma without brain metastases; however, the activity of dabrafenib plus trametinib has not been studied in active melanoma brain metastases. Here, we report results from the phase 2 COMBI-MB trial. Our aim was to build on the current body of evidence of targeted therapy in melanoma brain metastases through an evaluation of dabrafenib plus trametinib in patients with BRAFV600-mutant melanoma brain metastases.

METHODS

This ongoing, multicentre, multicohort, open-label, phase 2 study evaluated oral dabrafenib (150 mg twice per day) plus oral trametinib (2 mg once per day) in four patient cohorts with melanoma brain metastases enrolled from 32 hospitals and institutions in Europe, North America, and Australia: (A) BRAFV600E-positive, asymptomatic melanoma brain metastases, with no previous local brain therapy, and an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1; (B) BRAFV600E-positive, asymptomatic melanoma brain metastases, with previous local brain therapy, and an ECOG performance status of 0 or 1; (C) BRAFV600D/K/R-positive, asymptomatic melanoma brain metastases, with or without previous local brain therapy, and an ECOG performance status of 0 or 1; and (D) BRAFV600D/E/K/R-positive, symptomatic melanoma brain metastases, with or without previous local brain therapy, and an ECOG performance status of 0, 1, or 2. The primary endpoint was investigator-assessed intracranial response in cohort A in the all-treated-patients population. Secondary endpoints included intracranial response in cohorts B, C, and D. This study is registered with ClinicalTrials.gov, number NCT02039947.

FINDINGS

Between Feb 28, 2014, and Aug 5, 2016, 125 patients were enrolled in the study: 76 patients in cohort A; 16 patients in cohort B; 16 patients in cohort C; and 17 patients in cohort D. At the data cutoff (Nov 28, 2016) after a median follow-up of 8·5 months (IQR 5·5-14·0), 44 (58%; 95% CI 46-69) of 76 patients in cohort A achieved an intracranial response. Intracranial response by investigator assessment was also achieved in nine (56%; 95% CI 30-80) of 16 patients in cohort B, seven (44%; 20-70) of 16 patients in cohort C, and ten (59%; 33-82) of 17 patients in cohort D. The most common serious adverse events related to study treatment were pyrexia for dabrafenib (eight [6%] of 125 patients) and decreased ejection fraction (five [4%]) for trametinib. The most common grade 3 or worse adverse events, regardless of study drug relationship, were pyrexia (four [3%] of 125) and headache (three [2%]).

INTERPRETATION

Dabrafenib plus trametinib was active with a manageable safety profile in this melanoma population that was consistent with previous dabrafenib plus trametinib studies in patients with BRAFV600-mutant melanoma without brain metastases, but the median duration of response was relatively short. These results provide evidence of clinical benefit with dabrafenib plus trametinib and support the need for additional research to further improve outcomes in patients with melanoma brain metastases.

FUNDING

Novartis.

NRP-1 targeted and cargo-loaded exosomes facilitate simultaneous imaging and therapy of glioma in vitro and in vivo

Currently, glioma treatment is limited by two main factors: timely detection at onset or relapse and restriction of drugs by the blood-brain barrier (BBB) from entering the brain and influencing tumor growth. However, a safe BBB-traversing drug delivery system has brought new hope to glioma treatment. Exosomes have strong cargo-loading capacity and have the ability to cross the BBB. They can also be conferred with the ability for targeted delivery. Therefore, exosomes have great promise to be a targeted drug delivery vehicles. In this study, we firstly loaded superparamagnetic iron oxide nanoparticles (SPIONs) and curcumin (Cur) into exosomes and then conjugated the exosome membrane with neuropilin-1-targeted peptide (RGERPPR, RGE) by click chemistry to obtain glioma-targeting exosomes with imaging and therapeutic functions. When administered to glioma cells and orthotopic glioma models, we found that these engineered exosomes could cross the BBB smoothly and provided good results for targeted imaging and therapy of glioma. Furthermore, SPION-mediated magnetic flow hyperthermia (MFH) and Cur-mediated therapy also showed a potent synergistic antitumor effect. Therefore, the diagnostic and therapeutic effects on glioma were significantly improved, while reducing the side effects. We have designed a new type of glioma-targeting exosomes, which can carry nanomaterials and chemical agents for simultaneous diagnosis and treatment of glioma, thus providing a potential approach for improving the diagnosis and treatment effects of intracranial tumors.

Distribution of brain metastases

The number and site of brain metastases were identified on the computed tomographic scans of 288 patients. There was one brain metastasis in 49%, two in 21%, three in 13%, four in 6%, and five or more in 11% of scans. In patients with one metastasis, the posterior fossa was involved in 50% of patients when the primary tumor was pelvic (prostate or uterus) or gastrointestinal, but it was involved in only 10% of patients with other primary tumors. Hemispheral metastases preferred the anatomic "watershed areas" (29% of the brain surface contained 37% of the metastases), indicating that tumoral microemboli tend to lodge in the capillaries of the distal parts of the superficial arteries. The charts of 134 patients with brain metastases from a primary tumor originating outside the lung revealed that the incidence of lung and spine metastases was the same, whether the primary tumor was pelvic or gastrointestinal or from another site. These data suggest that the high incidence of subtentorial lesions in patients with pelvic and gastrointestinal primary tumors cannot be explained by arterial embolization alone, and that this peculiar distribution is probably not explained by seeding of the brain through Batson's plexus.

O-(2-[18F]fluoroethyl)-L-tyrosine PET combined with MRI improves the diagnostic assessment of cerebral gliomas

MRI is commonly used to determine the location and extent of cerebral gliomas. We investigated whether the diagnostic accuracy of MRI could be improved by the additional use of PET with the amino acid O-(2-[18F]fluoroethyl)-l-tyrosine (FET). In a prospective study, PET with FET and MRI was performed in 31 patients with suspected cerebral gliomas. PET and MRIs were co-registered and 52 neuronavigated tissue biopsies were taken from lesions with both abnormal MRI signal and increased FET uptake (match), as well as from areas with abnormal MR signal but normal FET uptake or vice versa (mismatch). Biopsy sites were labelled by intracerebral titanium pellets. The diagnostic performance for the identification of cellular tumour tissue was analysed for either MRI alone or MRI combined with FET PET using alternative free response receiver operating characteristic curves (ROCs). Histologically, 26 biopsy samples corresponded to cellular glioma tissue and 26 to peritumoral brain tissue. The diagnostic performance, as determined by the area under the ROC curve (Az), was Az = 0.80 for MRI alone and Az = 0.98 for the combined MRI and FET PET approach (P < 0.001). MRI yielded a sensitivity of 96% for the detection of tumour tissue but a specificity of only 53%, and combined use of MRI and FET PET yielded a sensitivity of 93% and a specificity of 94%. Combined use of MRI and FET PET in patients with cerebral gliomas significantly improves the identification of cellular glioma tissue and allows definite histological tumour diagnosis. Thus, our findings may have considerable impact on target selection for diagnostic biopsies as well as therapy planning.

Near real-time intraoperative brain tumor diagnosis using stimulated Raman histology and deep neural networks

Intraoperative diagnosis is essential for providing safe and effective care during cancer surgery1. The existing workflow for intraoperative diagnosis based on hematoxylin and eosin staining of processed tissue is time, resource and labor intensive2,3. Moreover, interpretation of intraoperative histologic images is dependent on a contracting, unevenly distributed, pathology workforce4. In the present study, we report a parallel workflow that combines stimulated Raman histology (SRH)5-7, a label-free optical imaging method and deep convolutional neural networks (CNNs) to predict diagnosis at the bedside in near real-time in an automated fashion. Specifically, our CNNs, trained on over 2.5 million SRH images, predict brain tumor diagnosis in the operating room in under 150 s, an order of magnitude faster than conventional techniques (for example, 20-30 min)2. In a multicenter, prospective clinical trial (n = 278), we demonstrated that CNN-based diagnosis of SRH images was noninferior to pathologist-based interpretation of conventional histologic images (overall accuracy, 94.6% versus 93.9%). Our CNNs learned a hierarchy of recognizable histologic feature representations to classify the major histopathologic classes of brain tumors. In addition, we implemented a semantic segmentation method to identify tumor-infiltrated diagnostic regions within SRH images. These results demonstrate how intraoperative cancer diagnosis can be streamlined, creating a complementary pathway for tissue diagnosis that is independent of a traditional pathology laboratory.

The Cyberknife: a frameless robotic system for radiosurgery

The Cyberknife is a unique instrument for performing frameless stereotactic radiosurgery. Rather than using rigid immobilization, the Cyberknife relies on an image-to-image correlation algorithm for target localization. Furthermore, the system utilizes a novel, light-weight, high-energy radiation source. The authors describe the technical specifications of the Cyberknife and summarize the initial clinical experience.

Experimental demonstration of noninvasive transskull adaptive focusing based on prior computed tomography scans

Developing minimally invasive brain surgery by high-intensity focused ultrasound beams is of great interest in cancer therapy. However, the skull induces strong aberrations both in phase and amplitude, resulting in a severe degradation of the beam shape. Thus, an efficient brain tumor therapy would require an adaptive focusing, taking into account the effects of the skull. In this paper, we will show that the acoustic properties of the skull can be deduced from high resolution CT scans and used to achieve a noninvasive adaptive focusing. Simulations have been performed with a full 3-D finite differences code, taking into account all the heterogeneities inside the skull. The set of signals to be emitted in order to focus through the skull can thus be computed. The complete adaptive focusing procedure based on prior CT scans has been experimentally validated. This could have promising applications in brain tumor hyperthermia but also in transcranial ultrasonic imaging.

GLP-1 receptor expression in human tumors and human normal tissues: potential for in vivo targeting

Peptide hormone receptors overexpressed in human tumors, such as somatostatin receptors, can be used for in vivo targeting for diagnostic and therapeutic purposes. A novel promising candidate in this field is the GLP-1 receptor, which was recently shown to be massively overexpressed in gut and lung neuroendocrine tumors--in particular, in insulinomas. Anticipating a major development of GLP-1 receptor targeting in nuclear medicine, our aim was to evaluate in vitro the GLP-1 receptor expression in a large variety of other tumors and to compare it with that in nonneoplastic tissues.

METHODS

The GLP-1 receptor protein expression was qualitatively and quantitatively investigated in a broad spectrum of human tumors (n=419) and nonneoplastic human tissues (n=209) with receptor autoradiography using (125)I-GLP-1(7-36)amide. Pharmacologic competition experiments were performed to provide proof of specificity of the procedure.

RESULTS

GLP-1 receptors were expressed in various endocrine tumors, with particularly high amounts in pheochromocytomas, as well as in brain tumors and embryonic tumors but not in carcinomas or lymphomas. In nonneoplastic tissues, GLP-1 receptors were present in generally low amounts in specific tissue compartments of several organs--namely, pancreas, intestine, lung, kidney, breast, and brain; no receptors were identified in lymph nodes, spleen, liver, or the adrenal gland. The rank order of potencies for receptor binding--namely, GLP-1(7-36)amide = exendin-4 >> GLP-2 = glucagon(1-29)--provided proof of specific GLP-1 receptor identification.

CONCLUSION

The GLP-1 receptors may represent a novel molecular target for in vivo scintigraphy and targeted radiotherapy for a variety of GLP-1 receptor-expressing tumors. For GLP-1 receptor scintigraphy, a low-background signal can be expected, on the basis of the low receptor expression in the normal tissues surrounding tumors.

Imaging tumor angiogenesis with contrast ultrasound and microbubbles targeted to alpha(v)beta3

BACKGROUND

Angiogenesis is a critical determinant of tumor growth and metastasis. We hypothesized that contrast-enhanced ultrasound (CEU) with microbubbles targeted to alpha(v)-integrins expressed on the neovascular endothelium could be used to image angiogenesis.

METHODS AND RESULTS

Malignant gliomas were produced in 14 athymic rats by intracerebral implantation of U87MG human glioma cells. On day 14 or day 28 after implantation, CEU was performed with microbubbles targeted to alpha(v)beta3 by surface conjugation of echistatin. CEU perfusion imaging with nontargeted microbubbles was used to derive tumor microvascular blood volume and blood velocity. Vascular alpha(v)-integrin expression was assessed by immunohistochemistry, and microbubble adhesion was characterized by confocal microscopy. Mean tumor size increased markedly from 14 to 28 days (2+/-1 versus 35+/-14 mm2, P<0.001). Tumor blood volume increased by approximately 35% from day 14 to day 28, whereas microvascular blood velocity decreased, especially at the central portions of the tumors. On confocal microscopy, alpha(v)beta3-targeted but not control microbubbles were retained preferentially within the tumor microcirculation. CEU signal from alpha(v)beta3-targeted microbubbles in tumors increased significantly from 14 to 28 days (1.7+/-0.4 versus 3.3+/-1.0 relative units, P<0.05). CEU signal from alpha(v)beta3-targeted microbubbles was greatest at the periphery of tumors, where alpha(v)-integrin expression was most prominent, and correlated well with tumor microvascular blood volume (r=0.86).

CONCLUSIONS

CEU with microbubbles targeted to alpha(v)beta3 can noninvasively detect early tumor angiogenesis. This technique, when coupled with changes in blood volume and velocity, may provide insights into the biology of tumor angiogenesis and be used for diagnostic applications.

Atypical and anaplastic meningiomas: radiology, surgery, radiotherapy, and outcome

Out of 936 primary intracranial meningiomas, 94.3% were histologically benign (grade I), 4.7% atypical (grade II), and 1.0% anaplastic (grade III); one recurrence was sarcomatous (grade IV). Meningiomas with histologic anaplasia (grades II-IV) occurred in 12% of the men, but only 4% of the women. Only 26% of atypical or anaplastic meningiomas appeared completely innocent on a computed tomography scan. Angiograms, usually showing a meningeal feeding artery, suggested meningioma when computed tomography scans did not. At 5 years after complete removal, the recurrence rate was only 3% (21% at 25 years) for benign meningiomas, but 38% for atypical ones, and 78% for anaplastic ones. The median times to recurrence were 7.5, 2.4, and 3.5 years, respectively. In spite of postoperative radiotherapy, four of five anaplastic meningiomas recurred.

Verb generation in patients with focal frontal lesions: a neuropsychological test of neuroimaging findings

What are the neural bases of semantic memory? Traditional beliefs that the temporal lobes subserve the retrieval of semantic knowledge, arising from lesion studies, have been recently called into question by functional neuroimaging studies finding correlations between semantic retrieval and activity in left prefrontal cortex. Has neuroimaging taught us something new about the neural bases of cognition that older methods could not reveal or has it merely identified brain activity that is correlated with but not causally related to the process of semantic retrieval? We examined the ability of patients with focal frontal lesions to perform a task commonly used in neuroimaging experiments, the generation of semantically appropriate action words for concrete nouns, and found evidence of the necessity of the left inferior frontal gyrus for certain components of the verb generation task. Notably, these components did not include semantic retrieval per se.

Vascular targeted nanoparticles for imaging and treatment of brain tumors

PURPOSE

Development of new therapeutic drug delivery systems is an area of significant research interest. The ability to directly target a therapeutic agent to a tumor site would minimize systemic drug exposure, thus providing the potential for increasing the therapeutic index.

EXPERIMENTAL DESIGN

Photodynamic therapy (PDT) involves the uptake of a sensitizer by the cancer cells followed by photoirradiation to activate the sensitizer. PDT using Photofrin has certain disadvantages that include prolonged cutaneous photosensitization. Delivery of nanoparticles encapsulated with photodynamic agent specifically to a tumor site could potentially overcome the drawbacks of systemic therapy. In this study, we have developed a multifunctional polymeric nanoparticle consisting of a surface-localized tumor vasculature targeting F3 peptide and encapsulated PDT and imaging agents.

RESULTS

The nanoparticles specifically bound to the surface of MDA-435 cells in vitro and were internalized conferring photosensitivity to the cells. Significant magnetic resonance imaging contrast enhancement was achieved in i.c. rat 9L gliomas following i.v. nanoparticle administration. Serial magnetic resonance imaging was used for determination of pharmacokinetics and distribution of nanoparticles within the tumor. Treatment of glioma-bearing rats with targeted nanoparticles followed by PDT showed a significant improvement in survival rate when compared with animals who received PDT after administration of nontargeted nanoparticles or systemic Photofrin.

CONCLUSIONS

This study reveals the versatility and efficacy of the multifunctional nanoparticle for the targeted detection and treatment of cancer.

Modified Criteria for Radiographic Response Assessment in Glioblastoma Clinical Trials

Radiographic endpoints including response and progression are important for the evaluation of new glioblastoma therapies. The current RANO criteria was developed to overcome many of the challenges identified with previous guidelines for response assessment, however, significant challenges and limitations remain. The current recommendations build on the strengths of the current RANO criteria, while addressing many of these limitations. Modifications to the current RANO criteria include suggestions for volumetric response evaluation, use contrast enhanced T1 subtraction maps to increase lesion conspicuity, removal of qualitative non-enhancing tumor assessment requirements, use of the post-radiation time point as the baseline for newly diagnosed glioblastoma response assessment, and "treatment-agnostic" response assessment rubrics for identifying pseudoprogression, pseudoresponse, and a confirmed durable response in newly diagnosed and recurrent glioblastoma trials.

Topographic anatomy and CT correlations in the untreated glioblastoma multiforme

To provide baseline information for the "local" therapy of the glioblastoma multiforme (GBM), whole-brain histological sections of 15 untreated GBM's were studied to determine the distribution of neoplastic cells. These findings were then compared with the computerized tomography (CT) scans in 11 cases in order to determine the extent to which the peripheral portion of the neoplasm can be estimated by the presence of a low-density area without contrast enhancement. The results of the histological study confirmed the marked heterogeneity of GBM's and disclosed a great variability in the geometry, extent, and character of the peripheral "infiltrating" margin. In spite of the widely held concept that glioblastomas are localized within 2 cm of the contrast-enhanced rim, there were three cases in this two-dimensional study in which this distance was exceeded, and it seems likely that three-dimensional reconstructions would have detected additional cases in which neoplastic cells extended beyond this arbitrary limit. Only three of the 15 GBM's were restricted to the distribution of one internal carotid or one vertebral artery. To the extent that the neoplasms in the present series are representative, this suggests that glioblastomas will be difficult to treat successfully by intra-arterial therapy using existing therapeutic agents. Correlations of histological sections with the CT scans revealed that the vast majority of the neoplastic tissue was contained within the contrast-enhancing and "peritumoral" areas of low density, but that in five cases fingers of neoplasm extended for short distances beyond the outer margin of the latter region. This indicates that the distribution of cells of a GBM cannot be inferred from CT images since the "peritumoral" area of low density can over- or underestimate the extent of the lesion.

Interactive Medical Image Segmentation Using Deep Learning With Image-Specific Fine Tuning

Convolutional neural networks (CNNs) have achieved state-of-the-art performance for automatic medical image segmentation. However, they have not demonstrated sufficiently accurate and robust results for clinical use. In addition, they are limited by the lack of image-specific adaptation and the lack of generalizability to previously unseen object classes (a.k.a. zero-shot learning). To address these problems, we propose a novel deep learning-based interactive segmentation framework by incorporating CNNs into a bounding box and scribble-based segmentation pipeline. We propose image-specific fine tuning to make a CNN model adaptive to a specific test image, which can be either unsupervised (without additional user interactions) or supervised (with additional scribbles). We also propose a weighted loss function considering network and interaction-based uncertainty for the fine tuning. We applied this framework to two applications: 2-D segmentation of multiple organs from fetal magnetic resonance (MR) slices, where only two types of these organs were annotated for training and 3-D segmentation of brain tumor core (excluding edema) and whole brain tumor (including edema) from different MR sequences, where only the tumor core in one MR sequence was annotated for training. Experimental results show that: 1) our model is more robust to segment previously unseen objects than state-of-the-art CNNs; 2) image-specific fine tuning with the proposed weighted loss function significantly improves segmentation accuracy; and 3) our method leads to accurate results with fewer user interactions and less user time than traditional interactive segmentation methods.

Noninvasive detection of therapeutic cytolytic T cells with 18F-FHBG PET in a patient with glioma

BACKGROUND

A 57-year-old man had been diagnosed with grade IV glioblastoma multiforme and was enrolled in a trial of adoptive cellular immunotherapy. The trial involved infusion of ex vivo expanded autologous cytolytic CD8+ T cells (CTLs), genetically engineered to express the interleukin 13 zetakine gene (which encodes a receptor protein that targets these T cells to tumor cells) and the herpes simplex virus 1 thymidine kinase (HSV1 tk) suicide gene, and PET imaging reporter gene.

INVESTIGATIONS

MRI, whole-body and brain PET scan with (18)F-radiolabelled 9-[4-fluoro-3-(hydroxymethyl)butyl]guanine ((18)F-FHBG) to detect CTLs that express HSV1 tk, and safety monitoring after injection of (18)F-FHBG.

DIAGNOSIS

MRI detected grade III-IV glioblastoma multiforme plus two tumors recurrences that developed after resection of the initial tumor.

MANAGEMENT

Surgical resection of primary glioblastoma tumor, enrollment in CTL therapy trial, reresection of glioma recurrences, infusion of approximately 1 x 10(9) CTLs into the site of tumor reresection, and (18)F-FHBG PET scan to detect infused CTLs.

Computerized tomographic and pathologic studies of the untreated, quiescent, and recurrent glioblastoma multiforme

Pathological findings in 20 cases of glioblastoma multiforme were correlated with clinical histories and computerized tomographic (CT) scans. This was done to define the neoplasm in three stages: before treatment, during remission, and during recurrence. The untreated lesions were markedly cellular neoplasms composed predominantly of small anaplastic cells. The radiographic central region of low density was necrosis, the enhancing rim was a cellular zone of viable neoplasm, and the perilesional low-density area was edema with infiltrating tumor. In these 20 cases, all of the identifiable neoplasms lay within the zone of peritumoral edema or contrast enhancement, although small anaplastic cells may have been present in more distant regions. The lesions in remission were remarkable for their minimal mass effect, discrete nature, extensive necrosis, and content of large bizarre glia. The large cells were confined to the original tumor bed and were consistent with neoplastic cells inactivated and immobilized by radio- and chemotherapy. These lesions were accurately localized by CT scanning. The recurrent lesions were heterogeneous, but most were formed of widely disseminated small anaplastic cells. The highly cellular regions of such lesions could be localized by CT scanning, but CT could not detect less cellular foci in the cerebrum, cerebellum, or brain stem. In one patient, the contrast-enhancing lesions of "recurrence," were foci of radionecrosis, underscoring the difficulty in distinguishing this entity from recurrent neoplasm.