PConv-UNet: Multi-scale pinwheel convolutions for breast ultrasound tumor segmentation

P1.06-04 Deep Learning-Based Segmentation of Mesothelioma on CT Scans: Application to Patient Scans Exhibiting Pleural Effusion

Hepatic epithelioid hemangioendothelioma (EHE) is a rare malignant tumor with unknown pathogenesis. Herein, we report a case of a hepatic EHE presenting synchronously with a hepatocellular carcinoma (HCC). To the best of our knowledge, this is the second case report of synchronous hepatic EHE and HCC. An 84-year-old man presented with back pain. During examination, a tumor in liver segment 3 was coincidentally detected. Tumor marker (carbohydrate antigen 19-9, alpha-fetoprotein, and protein induced by vitamin K absence or antagonist-II) levels were elevated. Contrast-enhanced computed tomography revealed perinodular enhancement in the arterial and portal phases. Another tumor was detected in liver segment 2, which was homogeneously enhanced in the arterial phase, followed by washout in the portal and late phases. Based on these imaging findings, we diagnosed the tumor in segment 3 as a solitary cholangiocellular carcinoma and the tumor in segment 2 as a solitary HCC. Lateral sectionectomy of the liver was performed. Microscopically, spindle-shaped and epithelioid cells were present in the tumor in segment 3. On immunohistochemistry, the tumor cells were positive for CD31 and CD34, focally positive for D2-40, and negative for AE1/AE3. Therefore, the tumor in segment 3 was ultimately diagnosed as an EHE and the tumor in segment 2 as a well-differentiated HCC. Preoperative diagnosis of EHE is difficult owing to the lack of specific findings. Intratumoral calcification, halo sign, and lollipop sign are occasionally found in EHE and are useful imaging findings for diagnosis. Clinical behavior is unpredictable, ranging from indolent growth to rapid progression. Clinical or pathological predictors of the course of EHE are urgently required.

A 60-year-old man was referred to our hospital for the evaluation and treatment of general malaise. Contrast-enhanced computed tomography detected sigmoid colon cancer that had invaded the bladder, multiple liver metastases, and a small intestinal tumor. Hartmann's procedure was performed, with partial bladder and small bowel resection. A pathological examination revealed that the patient had sigmoid colon cancer and a gastrointestinal stromal tumor. The biopsy findings of a tumor in segment 8 of the liver indicated the presence of adenocarcinoma, thereby indicating the origin of multiple liver metastases from sigmoid colon cancer. On chemotherapy, the tumors in liver segments 2/3 and 8 shrank. However, the tumor in segment 6 enlarged. Since radical resection of all metastatic liver tumors was possible, hepatectomy was performed 10months after the initial surgery. A pathological examination revealed that the tumors in segments 2/3, 4, and 8 were adenocarcinomas and the tumors in segments 4, 6, and 7 had originated from the gastrointestinal stromal tumor. This suggested the coexistence of liver metastases from sigmoid colon cancer and the gastrointestinal stromal tumor. In cases involving multiple primary tumors, it is necessary to consider the possible coexistence of multiple metastases from different primary tumors.

Background: Growing evidence supports the critical role of preexisting immune response in triple negative breast cancer (TNBC). However, there are limitations with current evaluation approaches: inability to functionally assess the type of immune infiltration with traditional pathologic evaluation and loss of spatial distribution in conventional high-plex gene or protein expression analyses from the whole tumor section. Here, we report the initial analysis of immune protein expression as a function of spatial distribution and clinical outcomes in TNBC samples. Methods: NanoString GeoMxTM Digital Spatial Profiling (DSP) was used to quantify 39 immune-related proteins in stromal and tumor segments from 44 TNBC samples from the FinXX trial. Samples were matched for patient characteristics, treatment arm (capecitabine vs. 5-fluorouracil), and outcome based on recurrence-free survival (RFS) with 22 samples from patients who recurred and 22 samples from patients with durable RFS. Regions of interest (ROIs) were selected based on expression of cytokeratin (tumor), CD45 (leukocytes), or CD68 (macrophages). Each ROI was segmented into tumor (pancytokeratin-positive area) and stroma (pancytokeratin-negative/nuclear SYTO13-positive area). The general linear model was used for statistical analysis of differential expression with RFS as a categorical variable (recur yes or no). Results: A total of 950 tumor and stroma segments were included in this initial analysis. In both tumor and stroma segments, over-expression of T cell activation markers (CD137, GITR) was associated with better outcome, whereas T cell markers (CD3, CD4, CD8) were not significantly associated with outcome. In tumor segments alone, improved outcome was significantly associated with increased protein expression [> 2-fold change (FC) at p<0.001] of CD56, PD-L2, HLA-DR, CD137, GITR, and CD40. In CD45-enriched stroma, improved outcome was associated with elevated expression (FC>2.0, p<0.001) of PD-L2; whereas durable RFS was associated with elevated PD-L2 and IDO1 expression in CD68-enriched stroma. In contrast, macrophage/dendritic cell markers CD68 and CD11c were not associated with outcome. In tumor cells adjacent to CD45-enriched stroma, durable RFS was associated with increased abundance (FC>2.0, p<0.001) of PD-L2, CD56, CD27, GITR, CD20, HLA-DR, and IDO1. Similarly, in tumor segment associated with CD68-enriched stroma, proteins associated with RFS included PD-L2, HLA-DR, CD56, GITR, and CD137. Among 39 immune function proteins, only elevated CTLA4 expression in CD68-enriched stroma was associated with recurrence (FC=0.55, p<0.001), whereas tumor-segment CTLA4 was associated with RFS. Conclusions: Using an in-depth analysis to precisely quantify the abundance of multiple immune function proteins in a spatially defined manner, we observed that PD-L2, IDO1, and T cell activation markers were robustly associated with durable RFS in both tumor and stromal segments. In contrast, MHC components (HLA-DR, beta-2-microglobulin), B cell markers (CD20), and NK cell markers (CD56) were strongly associated with favorable outcome in tumor but not in stromal segments. Our study highlights the conclusion that the immune landscape of TNBC is far too complex to be encompassed by any single molecular marker, and more detailed analyses of the DSP data reported here are ongoing with a view towards using quantitative multiplex analyses to refine our understanding of how therapeutic outcome is influenced by interactions among immune cells and between immune and tumor cells. Acknowledgements: Supported by the Breast Cancer Research Foundation (BCRF18-161), Bankhead Coley (6BC05 Florida Department of Health), 26.2 with Donna Foundation, US National Cancer Institute (CA15083), and the Canadian Cancer Center Citation Format: Saranya Chumsri, Douglas Hinerfeld, Jennifer M. Kachergus, Yaohua Ma, Heather A Brauer, Sarah Warren, Xue Wang, Torsten O. Nielsen, Karama Asleh, Heikki Joensuu, Edith A. Perez, E. A. Thompson. Digital spatial mapping of the immune landscape of triple negative breast cancer reveals novel features of immune-tumor cell interaction [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr PD5-06.

Segmentation of the primary tumor mass in neuroblastoma could aid radiologists by facilitating reproducible and objective quantification of the tumor's tissue composition and size. However, due to the heterogeneous nature of the tissue components of the neuroblastic tumor, ranging from low-attenuation necrosis to high-attenuation calcification, some of which possess strong similarities with adjacent nontumoral tissues in computed tomographic (CT) images, segmentation of the tumor is a difficult problem. In this context, landmarking methods are proposed to assist in the segmentation of neuroblastic tumors. Methods are proposed to identify and segment automatically the rib structure, the vertebral column, the spinal canal, the diaphragm, and the pelvic girdle. The use of the landmarks assisted in limiting the scope of the tumor segmentation process to the abdomen, and resulted in the reduction of the false-positive error rates by 26.9%, on the average, over 10 CT exams, and improved the result of segmentation of neuroblastic tumors.

Segmentation of tumor from magnetic resonance image (MRI) brain images is an emergent research area in the field of medical image segmentation. As segmentation of brain tumor plays an important role for necessary treatment and planning of tumor surgery. However, segmentation of the brain tumor is still a great challenge in clinics, specially automatic segmentation. In this paper we proposed hybridized multilevel thresholding and level set method for automatic segmentation of brain tumor. The innovation for this paper is to interface the initial segmentation from multilevel thresholding and extract a fine portrait using level set method with morphological operations. The results are compared with the existing method and also with radiologist manual segmentation which confirm the effectiveness of this hybridized paradigm for brain tumor segmentation.

BackgroundThe incidence of synchronous double primary hepatic cancers is extremely low. Cholangiolocellular carcinoma is also a rare disease.Case presentationA 58-year-old Japanese man was referred to our hospital for the treatment of multiple liver tumors revealed on computed tomography scans. He was hepatitis B and C positive and had undergone hemodialysis for 9 years due to chronic renal failure. Computed tomography scans revealed two hepatic tumors (each ≤ 1.0 cm in diameter) in segments 3 and 7. The preoperative diagnosis was multiple hepatocellular carcinomas. He underwent partial resections of his liver. The resected specimens revealed that the tumors in segments 3 and 7 were well-defined lesions of 8.0 mm and 14.0 mm, respectively. Pathological and immunohistochemical examinations confirmed the tumor in segment 3 to be a cholangiolocellular carcinoma and the tumor in segment 7 to be a hepatocellular carcinoma. Chronic inflammation could contribute to the different types of primary hepatic cancers. It may also give rise to various combinations of synchronous double primary hepatic cancer in patients with chronic liver disease.ConclusionsWe describe the sixth case of synchronous double primary hepatic cancers consisting of hepatocellular carcinoma and cholangiolocellular carcinoma in chronic damaged liver and review the literature. In patients with chronic liver disease, careful surveillance with imaging studies should be mandatory as various types of primary hepatic cancers could develop.

<p>Supplementary Figure S10: Representative images of CRC sample used in the assessment of spatial proteogenomic data quality. FFPE colorectal cancer (CRC) sections (A) (zoomed (B)) were stained with the GeoMx NGS Human Protein modules (147-plex), WTA, and antibodies against CD45 (Immune; left panel) and PanCK (Tumor; right panel). Supplementary Figure S11: ROI-to-ROI comparison of the proteogenomic data to the single analyte controls. CRC FFPE sections were stained with 147-plex GeoMx NGS human Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune). The Pearson’s R was calculated between each ROI from the proteogenomic assay against all ROIs in the single analyte (A) protein and (B) RNA controls. ROIs are colored according to region (immune or tumor). Protein targets with SNR ≥ 3 and top 400 expressing RNA targets with SNR ≥ 4 were used in the analysis. Supplementary Figure S12: Concordance between matching protein and RNA targets. For protein targets with SNR ≥ 3 and the respective RNA target with SNR ≥ 4, a pairwise scatterplot was generated to visualize the concordance between respective analytes. Pearson’s R calculations are shown in each plot. Supplementary Figure S13: Concordance between Proteogenomic RNA and Protein targets above background in immune segments of colorectal cancer (CRC). FFPE section of CRC stained with 147-plex GeoMx NGS Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune) using the proteogenomic workflow. Tumor and immune segments were selected based on PanCK and CD45 immunofluorescence, respectively. Pearson’s R was calculated between each detected protein target (SNR ≥ 3) and RNA targets (SNR ≥ 4) within the immune segment. Supplementary Figure S14: Concordance between Proteogenomic RNA and Protein targets above background in tumor segments of colorectal cancer (CRC). FFPE section of CRC stained with 147-plex GeoMx NGS Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune) using the proteogenomic workflow. Tumor and immune segments were selected based on PanCK and CD45 immunofluorescence, respectively. Pearson’s R was calculated between each detected protein target (SNR ≥ 3) and RNA targets (SNR ≥ 4) within the tumor segment. Supplementary Figure S15: Assessment of spatial proteogenomic performance on human NSCLC. NSCLC FFPE sections were stained with 147-plex GeoMx NGS human Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune). (A) Multiplexed protein and RNA characterization of NSCLC sample with representative colored and gray scaled images highlighting the segmentation of 300 μm circular ROIs into tumor (PanCK+) and immune (CD45+) enriched regions. Segments illuminated in white were collected, black regions were not. Protein and RNA counts were SNR transformed and protein targets with SNR ≥ 3 and RNA targets with SNR ≥ 4 were used in the analysis. (B) Combined volcano plot of protein and RNA expression in NSCLC. All immune segments were compared to all tumor segments for protein and RNA targets above background. A subset of differentially expressed genes are labeled with colors matching their analyte. (C) Unsupervised hierarchical clustering of detected RNA (left) and protein (right) targets for NSCLC.</p>

<p>Supplementary Figure S10: Representative images of CRC sample used in the assessment of spatial proteogenomic data quality. FFPE colorectal cancer (CRC) sections (A) (zoomed (B)) were stained with the GeoMx NGS Human Protein modules (147-plex), WTA, and antibodies against CD45 (Immune; left panel) and PanCK (Tumor; right panel). Supplementary Figure S11: ROI-to-ROI comparison of the proteogenomic data to the single analyte controls. CRC FFPE sections were stained with 147-plex GeoMx NGS human Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune). The Pearson’s R was calculated between each ROI from the proteogenomic assay against all ROIs in the single analyte (A) protein and (B) RNA controls. ROIs are colored according to region (immune or tumor). Protein targets with SNR ≥ 3 and top 400 expressing RNA targets with SNR ≥ 4 were used in the analysis. Supplementary Figure S12: Concordance between matching protein and RNA targets. For protein targets with SNR ≥ 3 and the respective RNA target with SNR ≥ 4, a pairwise scatterplot was generated to visualize the concordance between respective analytes. Pearson’s R calculations are shown in each plot. Supplementary Figure S13: Concordance between Proteogenomic RNA and Protein targets above background in immune segments of colorectal cancer (CRC). FFPE section of CRC stained with 147-plex GeoMx NGS Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune) using the proteogenomic workflow. Tumor and immune segments were selected based on PanCK and CD45 immunofluorescence, respectively. Pearson’s R was calculated between each detected protein target (SNR ≥ 3) and RNA targets (SNR ≥ 4) within the immune segment. Supplementary Figure S14: Concordance between Proteogenomic RNA and Protein targets above background in tumor segments of colorectal cancer (CRC). FFPE section of CRC stained with 147-plex GeoMx NGS Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune) using the proteogenomic workflow. Tumor and immune segments were selected based on PanCK and CD45 immunofluorescence, respectively. Pearson’s R was calculated between each detected protein target (SNR ≥ 3) and RNA targets (SNR ≥ 4) within the tumor segment. Supplementary Figure S15: Assessment of spatial proteogenomic performance on human NSCLC. NSCLC FFPE sections were stained with 147-plex GeoMx NGS human Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune). (A) Multiplexed protein and RNA characterization of NSCLC sample with representative colored and gray scaled images highlighting the segmentation of 300 μm circular ROIs into tumor (PanCK+) and immune (CD45+) enriched regions. Segments illuminated in white were collected, black regions were not. Protein and RNA counts were SNR transformed and protein targets with SNR ≥ 3 and RNA targets with SNR ≥ 4 were used in the analysis. (B) Combined volcano plot of protein and RNA expression in NSCLC. All immune segments were compared to all tumor segments for protein and RNA targets above background. A subset of differentially expressed genes are labeled with colors matching their analyte. (C) Unsupervised hierarchical clustering of detected RNA (left) and protein (right) targets for NSCLC.</p>

Automated Small Tumor Segmentation by a Template-Based Global Hierarchical Attention Method.

<p>Supplementary Figure S10: Representative images of CRC sample used in the assessment of spatial proteogenomic data quality. FFPE colorectal cancer (CRC) sections (A) (zoomed (B)) were stained with the GeoMx NGS Human Protein modules (147-plex), WTA, and antibodies against CD45 (Immune; left panel) and PanCK (Tumor; right panel). Supplementary Figure S11: ROI-to-ROI comparison of the proteogenomic data to the single analyte controls. CRC FFPE sections were stained with 147-plex GeoMx NGS human Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune). The Pearson’s R was calculated between each ROI from the proteogenomic assay against all ROIs in the single analyte (A) protein and (B) RNA controls. ROIs are colored according to region (immune or tumor). Protein targets with SNR ≥ 3 and top 400 expressing RNA targets with SNR ≥ 4 were used in the analysis. Supplementary Figure S12: Concordance between matching protein and RNA targets. For protein targets with SNR ≥ 3 and the respective RNA target with SNR ≥ 4, a pairwise scatterplot was generated to visualize the concordance between respective analytes. Pearson’s R calculations are shown in each plot. Supplementary Figure S13: Concordance between Proteogenomic RNA and Protein targets above background in immune segments of colorectal cancer (CRC). FFPE section of CRC stained with 147-plex GeoMx NGS Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune) using the proteogenomic workflow. Tumor and immune segments were selected based on PanCK and CD45 immunofluorescence, respectively. Pearson’s R was calculated between each detected protein target (SNR ≥ 3) and RNA targets (SNR ≥ 4) within the immune segment. Supplementary Figure S14: Concordance between Proteogenomic RNA and Protein targets above background in tumor segments of colorectal cancer (CRC). FFPE section of CRC stained with 147-plex GeoMx NGS Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune) using the proteogenomic workflow. Tumor and immune segments were selected based on PanCK and CD45 immunofluorescence, respectively. Pearson’s R was calculated between each detected protein target (SNR ≥ 3) and RNA targets (SNR ≥ 4) within the tumor segment. Supplementary Figure S15: Assessment of spatial proteogenomic performance on human NSCLC. NSCLC FFPE sections were stained with 147-plex GeoMx NGS human Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune). (A) Multiplexed protein and RNA characterization of NSCLC sample with representative colored and gray scaled images highlighting the segmentation of 300 μm circular ROIs into tumor (PanCK+) and immune (CD45+) enriched regions. Segments illuminated in white were collected, black regions were not. Protein and RNA counts were SNR transformed and protein targets with SNR ≥ 3 and RNA targets with SNR ≥ 4 were used in the analysis. (B) Combined volcano plot of protein and RNA expression in NSCLC. All immune segments were compared to all tumor segments for protein and RNA targets above background. A subset of differentially expressed genes are labeled with colors matching their analyte. (C) Unsupervised hierarchical clustering of detected RNA (left) and protein (right) targets for NSCLC.</p>

<p>Supplementary Figure S10: Representative images of CRC sample used in the assessment of spatial proteogenomic data quality. FFPE colorectal cancer (CRC) sections (A) (zoomed (B)) were stained with the GeoMx NGS Human Protein modules (147-plex), WTA, and antibodies against CD45 (Immune; left panel) and PanCK (Tumor; right panel). Supplementary Figure S11: ROI-to-ROI comparison of the proteogenomic data to the single analyte controls. CRC FFPE sections were stained with 147-plex GeoMx NGS human Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune). The Pearson’s R was calculated between each ROI from the proteogenomic assay against all ROIs in the single analyte (A) protein and (B) RNA controls. ROIs are colored according to region (immune or tumor). Protein targets with SNR ≥ 3 and top 400 expressing RNA targets with SNR ≥ 4 were used in the analysis. Supplementary Figure S12: Concordance between matching protein and RNA targets. For protein targets with SNR ≥ 3 and the respective RNA target with SNR ≥ 4, a pairwise scatterplot was generated to visualize the concordance between respective analytes. Pearson’s R calculations are shown in each plot. Supplementary Figure S13: Concordance between Proteogenomic RNA and Protein targets above background in immune segments of colorectal cancer (CRC). FFPE section of CRC stained with 147-plex GeoMx NGS Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune) using the proteogenomic workflow. Tumor and immune segments were selected based on PanCK and CD45 immunofluorescence, respectively. Pearson’s R was calculated between each detected protein target (SNR ≥ 3) and RNA targets (SNR ≥ 4) within the immune segment. Supplementary Figure S14: Concordance between Proteogenomic RNA and Protein targets above background in tumor segments of colorectal cancer (CRC). FFPE section of CRC stained with 147-plex GeoMx NGS Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune) using the proteogenomic workflow. Tumor and immune segments were selected based on PanCK and CD45 immunofluorescence, respectively. Pearson’s R was calculated between each detected protein target (SNR ≥ 3) and RNA targets (SNR ≥ 4) within the tumor segment. Supplementary Figure S15: Assessment of spatial proteogenomic performance on human NSCLC. NSCLC FFPE sections were stained with 147-plex GeoMx NGS human Protein modules, WTA, and antibodies against PanCK (Tumor) and CD45 (Immune). (A) Multiplexed protein and RNA characterization of NSCLC sample with representative colored and gray scaled images highlighting the segmentation of 300 μm circular ROIs into tumor (PanCK+) and immune (CD45+) enriched regions. Segments illuminated in white were collected, black regions were not. Protein and RNA counts were SNR transformed and protein targets with SNR ≥ 3 and RNA targets with SNR ≥ 4 were used in the analysis. (B) Combined volcano plot of protein and RNA expression in NSCLC. All immune segments were compared to all tumor segments for protein and RNA targets above background. A subset of differentially expressed genes are labeled with colors matching their analyte. (C) Unsupervised hierarchical clustering of detected RNA (left) and protein (right) targets for NSCLC.</p>

Segmentation of the tumor in neuroblastoma is complicated by the fact that the mass is almost always heterogeneous in nature; furthermore, viable tumor, necrosis, and normal tissue are often intermixed. Tumor definition and diagnosis require the analysis of the spatial distribution and Hounsfield unit (HU) values of voxels in computed tomography (CT) images, coupled with a knowledge of normal anatomy. Segmentation and analysis of the tissue composition of the tumor can assist in quantitative assessment of the response to therapy and in the planning of the delayed surgery for resection of the tumor. We propose methods to achieve 3-dimensional segmentation of the neuroblastic tumor. In our scheme, some of the normal structures expected in abdominal CT images are delineated and removed from further consideration; the remaining parts of the image volume are then examined for tumor mass. Mathematical morphology, fuzzy connectivity, and other image processing tools are deployed for this purpose. Expert knowledge provided by a radiologist in the form of the expected structures and their shapes, HU values, and radiological characteristics are incorporated into the segmentation algorithm. In this preliminary study, the methods were tested with 10 CT exams of four cases from the Alberta Children's Hospital. False-negative error rates of less than 12% were obtained in eight of 10 exams; however, seven of the exams had false-positive error rates of more than 20% with respect to manual segmentation of the tumor by a radiologist.

Segmentation of brain tumors has been found challenging throughout in the field of image processing. Different algorithms have been applied to the segmentation of solid or cystic tumors individually but little work has been done for solid cum cystic tumor. The papers reviewed in this article only deal with the case study of patients suffering from solid cum cystic brain tumor as this type of tumor is rarely found for the purpose of research. The research work conducted so far on this topic has been reviewed. The study begins with 2D (Two Dimensional) segmentation of tumor using MATLAB. It is then extended to study of slices of tumor and its volume calculation using open source software named 3D Slicer which represents the tumor in 3D. This software can intake the 2D slices and process them to give a combined 3D view. Various techniques are available in the software. According to the particular requirement an appropriate algorithm can be chosen. This paper gives a promising hierarchy for volume calculation of tumor and the three dimensional view. Further we can also find the position of tumor in the skull using the same software. This piece of work is a valuable guideline for the researchers interested in segmentation and three dimensional representations of different areas of human body. The models extracted out using the given algorithms can also be treated for matching and comparison of any future research. This will also aid surgeons and physicians in efficient analysis and reporting techniques.

BackgroundThe brain tumor is the growth of abnormal cells inside the brain. These cells can be grown into malignant or benign tumors. Segmentation of tumor from MRI images using image processing techniques started decades back. Image processing based brain tumor segmentation can be divided in to three categories conventional image processing methods, Machine Learning methods and Deep Learning methods. Conventional methods lacks the accuracy in segmentation due to complex spatial variation of tumor. Machine Learning methods stand as a good alternative to conventional methods. Methods like SVM, KNN, Fuzzy and a combination of either of these provide good accuracy with reasonable processing speed. The difficulty in processing the various feature extraction methods and maintain accuracy as per the medical standards still exist as a limitation for machine learning methods. In Deep Learning features are extracted automatically in various stages of the network and maintain accuracy as per the medical standards. Huge database requirement and high computational time is still poses a problem for deep learning. To overcome the limitations specified above we propose an unsupervised dual autoencoder with latent space optimization here. The model require only normal MRI images for its training thus reducing the huge tumor database requirement. With a set of normal class data, an autoencoder can reproduce the feature vector into an output layer. This trained autoencoder works well with normal data while it fails to reproduce an anomaly to the output layer. But a classical autoencoder suffer due to poor latent space optimization. The Latent space loss of classical autoencoder is reduced using an auxiliary encoder along with the feature optimization based on singular value decomposition (SVD). The patches used for training are not traditional square patches but we took both horizontal and vertical patches to keep both local and global appearance features on the training set. An Autoencoder is applied separately for learning both horizontal and vertical patches. While training a logistic sigmoid transfer function is used for both encoder and decoder parts. SGD optimizer is used for optimization with an initial learning rate of .001 and the maximum epochs used are 4000. The network is trained in MATLAB 2018a with a processor capacity of 3.7 GHz with NVIDIA GPU and 16 GB of RAM.ResultsThe results are obtained using a patch size of 16 × 64, 64 × 16 for horizontal and vertical patches respectively. In Glioma images tumor is not grown from a point rather it spreads randomly. Region filling and connectivity operations are performed to get the final tumor segmentation. Overall the method segments Meningioma better than Gliomas. Three evaluation metrics are considered to measure the performance of the proposed system such as Dice Similarity Coefficient, Positive Predictive Value, and Sensitivity.ConclusionAn unsupervised method for the segmentation of brain tumor from MRI images is proposed here. The proposed dual autoencoder with SVD based feature optimization reduce the latent space loss in the classical autoencoder. The proposed method have advantages in computational efficiency, no need of huge database requirement and better accuracy than machine learning methods. The method is compared Machine Learning methods Like SVM, KNN and supervised deep learning methods like CNN and commentable results are obtained.