SU-D-218-01: Support Vector Machine Tissue Classification of Multiparametric MRI Tumor Data.

PURPOSE: To test the accuracy of Support Vector Machines in the classification of Glioblastoma Multiforme tumor voxels usingmultiparametric MRI data. METHODS: Various MRI scans were collected from patients with recurrent GBM. Each scan session collected post-contrastT1(+C T1), T2, diffusion, perfusion, and multi-echo hypoxia images. Thediffusion-weighted images were converted to Apparent Diffusion Coefficient (ADC) maps. The perfusion images were corrected for leakage and represented as corrected rCBV maps. All of these scans were co-registered to each other, giving an input matrix to our support vector machine consisting of roughly 13,000 voxels, each with 5 feature values (+C T1, T2, ADC, rCBVcorrected, delta T2*). The SVM was then trained usingradiologist confirmed labels for 'cyst', 'tumor', and 'normal tissue'. These labels were obtained using longitudinal data as well as clinical scans, and tested on new data to determine the accuracy of classifying tumor and cyst voxels. RESULTS: The tumor model resulted in a specificity of 0.9841 and a sensitivity of 0.7498. The cyst model resulted in a specificity of 0.9825 anda sensitivity of 0.9414. Both models showed improvement with increasing features. CONCLUSIONS: These results show that the SVM is capable of classifying tumor and cyst voxels in a single case study. Aside from optimizing the current tumor model, future work will focus on the potential for the SVM to help in early detection of recurrence. In order to achieve this,we will need to test SVMs across patients. If we truly aim for early detection, then we need to prove that the algorithm can be trained on a pool of subjects with recurrent disease, and then test that model on a new patient before recurrence is obvious.

Sound localization acuity and its relation to vision in large and small fruit-eating bats: II. Non-echolocating species, Eidolon helvum and Cynopterus brachyotis.

Passive sound-localization acuity for 100-msec noise bursts was determined behaviorally for two species of non-echolocating bats: the Straw-colored fruit bat, Eidolon helvum, a large frugivore, and the Dog-faced fruit bat, Cynopterus brachyotis, a small frugivore. The mean minimum audible angle for two E. helvum was 11.7 degrees, and for two C. brachyotis was 10.5 degrees. This places their passive sound-localization acuity near the middle of the range for echolocating bats as well as the middle of the range for other mammals. Sound-localization acuity varies widely among mammals, and the best predictor of this auditory function remains the width of the field of best vision (r=.89, p<.0001). Among echolocating and non-echolocating bats, as well as among other mammals, the use of hearing to direct the eyes to the source of a sound still appears to serve as an important selective factor for sound localization. Absolute visual acuity and the magnitude of the binaural locus cues available to a species remain unreliable predictors of sound-localization acuity.

Sound-localization acuity and its relation to vision in large and small fruit-eating bats: I. Echolocating species, Phyllostomus hastatus and Carollia perspicillata.

Passive sound-localization acuity for 100-ms noise bursts was determined behaviorally for two species of bats: Phyllostomus hastatus, a large bat that eats fruit and vertebrates, and Carollia perspicillata, a small species that eats fruit and nectar. The mean minimum audible angle for two P. hastatus was 9 degrees , and that for two C. perspicillata was 14.8 degrees . This places their passive sound-localization acuity near the middle of the range for mammals. Sound localization varies widely among mammals and the best predictor of a species' acuity remains the width of the field of best vision (r=.89, p<.0001). The five echolocating bats that have been tested do not deviate from this relationship suggesting that despite their specialization for echolocation, the use of hearing to direct the eyes to the source of a sound still serves as an important selective factor for sound localization.

Hearing in large (Eidolon helvum) and small (Cynopterus brachyotis) non-echolocating fruit bats.

Comparing the hearing abilities of echolocating and non-echolocating bats can provide insight into the effect of echolocation on more basic hearing abilities. Toward this end, we determined the audiograms of two species of non-echolocating bats, the straw-colored fruit bat (Eidolon helvum), a large (230-350 g) African fruit bat, and the dog-faced fruit bat (Cynopterus brachyotis), a small (30-45 g) bat native to India and Southeast Asia. A conditioned suppression/avoidance procedure with a fruit juice reward was used for testing. At 60 dB SPL, the hearing range of E. helvum extends from 1.38 to 41 kHz with best sensitivity at 8k Hz; the hearing range of C. brachyotis extends from 2.63 to 70 kHz with best sensitivity at 10 kHz. As with all other bats tested so far, neither species was able to hear below 500 Hz, suggesting that they may not use a time code for perceiving pitch. Comparison of the high-frequency hearing abilities of echolocating and non-echolocating bats suggests that the use of laryngeal echolocation has resulted in additional selective pressure to hear high frequencies. However, the typical high-frequency sensitivity of small non-echolocating mammals would have been sufficient to support initial echolocation in the early evolution of bats, a finding that supports the possibility of multiple origins of echolocation.

Audiograms of five species of rodents: implications for the evolution of hearing and the perception of pitch.

Behavioral audiograms were determined for five species of rodents: groundhog (Marmota monax), chipmunk (Tamias striatus), Darwin's leaf-eared mouse (Phyllotis darwinii), golden hamster (Mesocricetus auratus), and Egyptian spiny mouse (Acomys cahirinus). The high-frequency hearing of these animals was found to vary inversely with interaural distance, a typical mammalian pattern. With regard to low-frequency hearing, the animals fell into two groups: those with extended low-frequency hearing (chipmunks, groundhogs, and hamsters hear below 100 Hz) and those with restricted low-frequency hearing (spiny and leaf-eared mice do not hear appreciably below 1 kHz). An analysis of mammalian hearing reveals that the distribution of low-frequency hearing limits is bimodal with the two distributions separated by a gap from 125 to 500 Hz. The correspondence of this dichotomy with studies of temporal coding raises the possibility that mammals that do not hear below 500 Hz do not use temporal encoding for the perception of pitch.

Sound localization in a new-world frugivorous bat, Artibeus jamaicensis: acuity, use of binaural cues, and relationship to vision.

Passive sound-localization acuity and its relationship to vision were determined for the echolocating Jamaican fruit bat (Artibeus jamaicensis). A conditioned avoidance procedure was used in which the animals drank fruit juice from a spout in the presence of sounds from their right, but suppressed their behavior, breaking contact with the spout, whenever a sound came from their left, thereby avoiding a mild shock. The mean minimum audible angle for three bats for a 100-ms noise burst was 10 degrees-marginally superior to the 11.6 degrees threshold for Egyptian fruit bats and the 14 degrees threshold for big brown bats. Jamaican fruit bats were also able to localize both low- and high-frequency pure tones, indicating that they can use both binaural phase- and intensity-difference cues to locus. Indeed, their ability to use the binaural phase cue extends up to 6.3 kHz, the highest frequency so far for a mammal. The width of their field of best vision, defined anatomically as the width of the retinal area containing ganglion-cell densities at least 75% of maximum, is 34 degrees. This value is consistent with the previously established relationship between vision and hearing indicating that, even in echolocating bats, the primary function of passive sound localization is to direct the eyes to sound sources.

Sound localization in an Old-World fruit bat (Rousettus aegyptiacus): acuity, use of binaural cues, and relationship to vision.

The passive sound-localization acuity of Egyptian fruit bats (Rousettus aegyptiacus) was determined using a conditioned-avoidance procedure. The mean minimum audible angle for left-right discrimination for 3 bats was 11.6 degrees--very near the mean for terrestrial mammals. The bats also were able to localize low- and high-frequency pure tones, indicating that they can use both binaural phase-difference and binaural intensity-difference cues to localize sound. Moreover, they were able to use the binaural phase-difference cue up to at least 5.6 kHz, which is higher than other mammals yet tested. The width of the Egyptian fruit bats' field of best vision was 27 degrees. This value is consistent with the hypothesis that the role of passive sound localization is to direct the eyes for visual scrutiny of sound sources. Thus, the passive localization abilities of these echolocating megachiropteran fruit bats do not deviate from the patterns established for nonecholocating mammals.

Hearing in a megachiropteran fruit bat (Rousettus aegyptiacus).

The Egyptian fruit bat (Rousettus aegyptiacus) is one of the few megachiropteran bats capable of echolocation. However, it uses rudimentary tongue clicks rather than laryngeally produced echo calls. We determined the audiogram of 2 bats using a conditioned avoidance procedure with fruit puree reward. At an intensity of 60 dB sound pressure level, the bats' hearing extended from 2.25 kHz to 64 kHz, with a region of good sensitivity between 8 kHz and 45 kHz. A dip in sensitivity at 32 kHz appears to be due to pinna directionality. The hearing of Egyptian fruit bats is typical for a mammal of that size and is not as limited as previously reported. Methodological issues, specifically training an animal to listen for low-intensity signals and imposing a significant cost for failing to report signals (i.e., misses), are discussed as the basis for the discrepancy between our results and earlier reports.

Passive sound-localization ability of the big brown bat (Eptesicus fuscus).

The passive sound-localization ability (i.e. minimum audible angle) of the big brown bat, Eptesicus fuscus, was determined using a conditioned avoidance procedure in which the animals were trained to discriminate left sounds from right sounds. The mean threshold of three bats for a 100-ms broadband noise burst was 14 degrees, a value that is about average for mammals. A similar threshold of 15 degrees was obtained for one animal when it was retested with one of its own recorded echolocation calls as the stimulus. The two bats tested on pure-tone localization were able to localize high-frequency, but not low-frequency tones, even when a low-frequency tone was amplitude modulated, a result indicating that these bats are not able to use binaural time-difference cues for localization. Finally, given the width of the bat's field of best vision, as determined by a count of its ganglion-cell density, its sound-localization acuity is consistent with the hypothesis that the role of passive sound localization is to direct the eyes to the source of a sound.

Audiogram of the big brown bat (Eptesicus fuscus).

The audiograms of three big brown bats (Eptesicus fuscus) were determined using a conditioned avoidance procedure. The average audiogram ranged from 0.850 kHz at 106 dB to 120 kHz at 83 dB SPL, with a best threshold of 7 dB at 20 kHz and a distinct decrease in sensitivity at 45 kHz. The results confirm those of a previous study by Dalland (1965a) that the big brown bat has good high-frequency hearing coupled with poor low-frequency hearing. Comparative analysis suggests that the bat's good high-frequency hearing initially evolved for passive sound localization and that it was later coopted for use in echolocation. In addition, the restricted low-frequency hearing of the big brown bat is typical of mammals with good high-frequency hearing.

Sound localization in chinchillas, III: Effect of pinna removal.

The ability of chinchillas to make left/right, front/back, and vertical locus discriminations was determined before and after surgical removal of the pinnae. The animals were tested behaviorally using a conditioned avoidance procedure. In the left/right localization tests, removal of both pinnae had no effect on localization acuity for broadband noise but did result in a small decrement in performance when localizing low-pass filtered noise. In the front/back localization tests, removal of a single pinna resulted in a small but consistent decrement in performance when the sound sources were located in the hemifield on the same side as the intact pinna, and a greater decrement when the sound sources were located in the hemifield on the side of the missing pinna; removal of both pinnae resulted in the largest decrement in performance. Finally, vertical localization acuity and performance when localizing low-pass filtered noise were greatly impaired following removal of both pinnae. These results demonstrate the importance of the pinnae in performing front/back and vertical localization tasks in which binaural cues are not available.

Sound localization in chinchillas. II. Front/back and vertical localization.

The ability of chinchillas to make front/back and vertical locus discriminations was examined behaviorally using a conditioned avoidance procedure. Their minimum audible angle for localizing single broadband noise bursts was 36 degrees for front/back localization and 23 degrees for vertical localization. Sound localization tests using filtered noise demonstrated that the signal must contain high frequencies in order for chinchillas to make front/back and vertical locus judgements and that frequencies in their highest audible octave (i.e., above 16 kHz) contribute to localization. These results support the view that a major selective advantage of high-frequency hearing in mammalian evolution was its utility for monaural as well as binaural sound localization.

Sound localization in chinchillas. I: Left/right discriminations.

The ability of chinchillas to localize sound was examined behaviorally using a conditioned avoidance procedure in which the animals were trained to discriminate left from right sound sources. Their minimum audible angle was 15.6 degrees for 100-ms broadband noise making them one of the more accurate rodents, although they are not as accurate as primates and carnivores. Thresholds obtained for filtered noise stimuli demonstrated that chinchillas are equally accurate in localizing either low- or high-frequency noise. Further, they are able to use both interaural phase-difference and interaural intensity-difference cues as demonstrated by their ability to localize both low- and high-frequency pure tones. Finally, analysis of the chinchilla retina supports the hypothesis that the role of auditory localization in directing the eyes to sound sources played a role in the evolution of auditory spatial perception.