Abstract— Conventional MLEM and OSEM algorithms used in SPECT Tc-99m sestamibi scintimammography produce hot spot artifacts (HSA). We investigated a suitable modification of MLEM and OSEM algorithms needed to reduce HSA. Patients with suspicious breast lesions were administered 10 mCi of Tc- 99m sestamibi and SPECT scans with patients in prone position with uncompressed breasts were acquired. In addition, to simulate breast lesions, some patients were imaged with a number of breast skin markers each containing 1 μCi of Tc-99m. We modified MLEM and OSEM algorithms by removing from the backprojection step the rays that traverse the periphery of the support region on the way to a detector bin when their path length trough this region is shorter than some preset critical length. Such very short paths result in a very low projection counts contributed to the detector bin and this in turn gives rise to a overestimation of the activity in the peripheral voxels in the backprojection step, thus creating HSA. We analyzed the breast- lesion contrast and suppression of HSA in the images reconstructed using conventional and modified MLEM and OSEM algorithms vs. critical path length (CPL). For CPL ≥ 0.01 pixel size, we observed improved breast-lesion contrast and lower noise in the images reconstructed, and a very significant Manuscript received November 12, 2004. A. Krol is with Department of Radiology, SUNY Upstate Medical University, 750 E. Adams. St., Syracuse NY 13210, USA (telephone: 315-464- 7054, e-mail: krola@upstate.edu). D. H. Feiglin is with Department of Radiology, SUNY Upstate Medical University, 750 E. Adams. St., Syracuse NY 13210, USA (telephone: 315-464- 031, e-mail: feiglind@upstate.edu). W. Lee is with Department of Radiology, SUNY Upstate Medical University, 750 E. Adams. St., Syracuse NY 13210, USA (telephone: 315-464- 7031, e-mail: wlee@visionexplore.com). V. R. Kunniyur is with Department of Electrical Engineering and Computer Science, Syracuse University, Syracuse, NY, USA (telephone: 315- 464-7031, e-mail: vrkunniy@ecs.syr.edu). K. R. Gangal is with Department of Electrical Engineering and Computer Science, Syracuse University, Syracuse, NY, USA (telephone: 315-464-7031, e-mail: krgangal@ecs.syr.edu). D. A. Karczewski is with Department of Radiology, SUNY Upstate Medical University, 750 E. Adams. St., Syracuse NY 13210, USA (telephone: 315-464-7031, e-mail: karczewd@upstate.edu). F. D. Thomas is with Department of Radiology, SUNY Upstate Medical University, 750 E. Adams. St., Syracuse NY 13210, USA (telephone: 315-464- 7031, e-mail: thomasd@upstate.edu). I. L. Coman is with Department of Computer Science and Mathematics, Ithaca College, Ithaca, NY, USA (telephone: 607-274-5704, e-mail: icoman@ithaca.edu). E. D. Lipson is with Department of Physics, Syracuse University, Syracuse, NY , USA (telephone: 315-443-9107, e-mail: edlipson@syr.edu). reduction of HSA in the maximum intensity projection (MIP) images I. INTRODUCTION C-99m sestamibi scintimammography is a useful tool for reducing the number of breast biopsies due to its very high negative predictive value (NPV) of 95% [1], as compared to screening x-ray mammography with positive predictive value (PPV) in the 15–40% range. Whereas scintimammography is usually performed in a planar acquisition mode, we explore here the application of a tomographic approach (SPECT) to scintimammography. II. MATERIALS AND METHODS Patients with suspicious breast lesions were administered 10 mCi of Tc-99m sestamibi. SPECT Tc-99m sestamibi scintimammography (90 views, 30 sec/view, parallel-hole high-resolution collimator) was acquired on a dual-head gamma camera (E.Cam, Siemens). In addition, in order to simulate hot breast lesions, some patients were imaged with a number of breast skin markers each containing 1 μCi of Tc- 99m. The images were reconstructed using a maximum-likelihood expectation-maximization (MLEM) algorithm [2-3] in its ordered subset version [4-5]. We performed fully-3D reconstruction with resolution and attenuation modeling. We used our of version of an MLEM algorithm [6]: λ k n +1 = λ k n 1 c ik i ∈S 0 c ik ⋅ γ iJ i Y i λ m n ⋅ c im ⋅ γ iJ i m∈P i + c ik (1 - γ iJ i ) i ∈S 0 . (1) The symbols are as follows: i projection subscript J i number of pixels in the ray I j pixel subscript (j < J i ) P i set of pixels contributing to projection i S 0 subset of the projection bins corresponding to a particular set of views Maximum-Likelihood Expectation-Maximization Algorithm for Improved Clinical SPECT Scintimammography Andrzej Krol, Member, IEEE, David H. Feiglin, Wei Lee, Vikram R. Kunniyur, Kedar R. Gangal, Ioana L. Coman, Member, IEEE, Edward D. Lipson, Member, IEEE, Deborah A. Karczewski, F. Deaver Thomas T 0-7803-8701-5/04/$20.00 (C) 2004 IEEE 0-7803-8700-7/04/$20.00 (C) 2004 IEEE 3523